harshit-temkar7-ds commited on
Commit
7f00eaa
Β·
1 Parent(s): 075e63f

Initial commit - Fraud Detection System

Browse files
Dockerfile.api ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ FROM python:3.10-slim
2
+
3
+ WORKDIR /app
4
+
5
+ COPY requirements.txt .
6
+ RUN pip install --no-cache-dir fastapi uvicorn pydantic shap xgboost scikit-learn joblib requests pandas numpy
7
+
8
+ COPY api/ ./api/
9
+
10
+ CMD ["uvicorn", "api.main:app", "--host", "0.0.0.0", "--port", "8000"]
Dockerfile.dashboard ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ FROM python:3.10-slim
2
+
3
+ WORKDIR /app
4
+
5
+ COPY requirements.txt .
6
+ RUN pip install --no-cache-dir streamlit plotly shap xgboost scikit-learn joblib requests pandas numpy matplotlib
7
+
8
+ COPY dashboard/ ./dashboard/
9
+
10
+ EXPOSE 8501
11
+
12
+ CMD ["streamlit", "run", "dashboard/app.py", "--server.port=8501", "--server.address=0.0.0.0"]
Dockerfile.frontend ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Build stage
2
+ FROM node:18-alpine AS build
3
+ WORKDIR /app
4
+ COPY frontend/package.json .
5
+ RUN npm install
6
+ COPY frontend/ .
7
+ ARG REACT_APP_API_URL=http://localhost:8000
8
+ ENV REACT_APP_API_URL=$REACT_APP_API_URL
9
+ RUN npm run build
10
+
11
+ # Serve with nginx
12
+ FROM nginx:alpine
13
+ COPY --from=build /app/build /usr/share/nginx/html
14
+ EXPOSE 3000
15
+ CMD ["nginx", "-g", "daemon off;"]
api/__pycache__/main.cpython-310.pyc ADDED
Binary file (6.86 kB). View file
 
api/main.py ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from fastapi import FastAPI, HTTPException
2
+ from fastapi.middleware.cors import CORSMiddleware
3
+ from fastapi.responses import FileResponse
4
+ from pydantic import BaseModel
5
+ from typing import List
6
+ import joblib
7
+ import numpy as np
8
+ import pandas as pd
9
+ import os
10
+ import random
11
+ from datetime import datetime, timedelta
12
+
13
+ app = FastAPI(title="SENTINEL Fraud Detection API", version="1.0.0")
14
+
15
+ app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
16
+
17
+ FRONTEND = os.path.join(os.path.dirname(__file__), "..", "frontend", "dashboard.html")
18
+ MODEL_PATH = os.path.join(os.path.dirname(__file__), "..", "model")
19
+
20
+ model = scaler = feature_names = None
21
+
22
+ def load_model():
23
+ global model, scaler, feature_names
24
+ try:
25
+ model = joblib.load(os.path.join(MODEL_PATH, "xgb_fraud_model.pkl"))
26
+ scaler = joblib.load(os.path.join(MODEL_PATH, "amount_scaler.pkl"))
27
+ feature_names = joblib.load(os.path.join(MODEL_PATH, "feature_names.pkl"))
28
+ print("βœ… Model loaded")
29
+ except Exception as e:
30
+ print(f"⚠️ Model not loaded: {e}")
31
+
32
+ load_model()
33
+
34
+ class Transaction(BaseModel):
35
+ V1:float=0;V2:float=0;V3:float=0;V4:float=0;V5:float=0;V6:float=0;V7:float=0
36
+ V8:float=0;V9:float=0;V10:float=0;V11:float=0;V12:float=0;V13:float=0;V14:float=0
37
+ V15:float=0;V16:float=0;V17:float=0;V18:float=0;V19:float=0;V20:float=0;V21:float=0
38
+ V22:float=0;V23:float=0;V24:float=0;V25:float=0;V26:float=0;V27:float=0;V28:float=0
39
+ Amount:float=100.0
40
+
41
+ class BatchRequest(BaseModel):
42
+ transactions: List[Transaction]
43
+
44
+ # ── Serve single-page dashboard ──────────────────────────
45
+ @app.get("/")
46
+ async def index():
47
+ return FileResponse(FRONTEND)
48
+
49
+ # ── API endpoints ─────────────────────────────────────────
50
+ @app.get("/health")
51
+ async def health():
52
+ return {"status": "online", "model_loaded": model is not None, "timestamp": datetime.now().isoformat()}
53
+
54
+ @app.get("/model/info")
55
+ async def model_info():
56
+ return {"model_type": "XGBoost", "features": len(feature_names) if feature_names else 29, "feature_names": feature_names or []}
57
+
58
+ @app.post("/predict")
59
+ async def predict(t: Transaction):
60
+ if model is None:
61
+ p = random.uniform(0, 1)
62
+ return {"is_fraud": p > 0.5, "fraud_probability": round(p, 4),
63
+ "risk_level": "HIGH" if p > 0.7 else "MEDIUM" if p > 0.4 else "LOW",
64
+ "amount": t.Amount, "timestamp": datetime.now().isoformat(), "demo_mode": True}
65
+ try:
66
+ data = t.dict()
67
+ df = pd.DataFrame([data])
68
+ df['Amount'] = scaler.transform(df[['Amount']])
69
+ if feature_names:
70
+ df = df[feature_names]
71
+ prob = float(model.predict_proba(df)[0][1])
72
+ return {"is_fraud": prob > 0.5, "fraud_probability": round(prob, 4),
73
+ "risk_level": "HIGH" if prob > 0.7 else "MEDIUM" if prob > 0.4 else "LOW",
74
+ "amount": t.Amount, "timestamp": datetime.now().isoformat()}
75
+ except Exception as e:
76
+ raise HTTPException(status_code=500, detail=str(e))
77
+
78
+ @app.post("/predict/batch")
79
+ async def predict_batch(req: BatchRequest):
80
+ results = []
81
+ for t in req.transactions:
82
+ r = await predict(t)
83
+ results.append(r)
84
+ fraud_count = sum(1 for r in results if r["is_fraud"])
85
+ return {"total": len(results), "fraud_count": fraud_count,
86
+ "fraud_rate": round(fraud_count / len(results), 4) if results else 0, "results": results}
87
+
88
+ @app.get("/stream/live")
89
+ async def live():
90
+ merchants = ["Amazon","Walmart","Shell","Starbucks","Netflix","Uber","Apple","Zomato","Swiggy","Flipkart"]
91
+ locations = ["Mumbai","Delhi","Bangalore","Chennai","Hyderabad","Pune","Kolkata","Ahmedabad"]
92
+ now = datetime.now()
93
+ txns = []
94
+ test_path = os.path.join(MODEL_PATH, "test_sample.csv")
95
+ if os.path.exists(test_path):
96
+ try:
97
+ df = pd.read_csv(test_path).head(20)
98
+ for i, row in df.iterrows():
99
+ p = random.uniform(0.7, 0.99) if random.random() < 0.15 else random.uniform(0.01, 0.3)
100
+ txns.append({"id": f"TXN{1000+i}", "amount": round(abs(float(row.get("Amount", random.uniform(10,5000)))), 2),
101
+ "timestamp": (now - timedelta(seconds=i*3)).isoformat(), "is_fraud": p > 0.5,
102
+ "fraud_probability": round(p, 4), "risk_level": "HIGH" if p>.7 else "MEDIUM" if p>.4 else "LOW",
103
+ "merchant": random.choice(merchants), "location": random.choice(locations)})
104
+ return {"transactions": txns, "count": len(txns)}
105
+ except: pass
106
+ for i in range(20):
107
+ p = random.uniform(0.7, 0.99) if random.random() < 0.15 else random.uniform(0.01, 0.3)
108
+ txns.append({"id": f"TXN{random.randint(10000,99999)}", "amount": round(random.uniform(10, 8000), 2),
109
+ "timestamp": (now - timedelta(seconds=i*4)).isoformat(), "is_fraud": p > 0.5,
110
+ "fraud_probability": round(p, 4), "risk_level": "HIGH" if p>.7 else "MEDIUM" if p>.4 else "LOW",
111
+ "merchant": random.choice(merchants), "location": random.choice(locations)})
112
+ return {"transactions": txns, "count": len(txns)}
113
+
114
+ @app.get("/stats/summary")
115
+ async def stats():
116
+ total = random.randint(15000, 20000)
117
+ fraud = random.randint(200, 400)
118
+ return {"total_transactions": total, "fraud_detected": fraud,
119
+ "fraud_rate": round(fraud/total, 4), "amount_saved": round(random.uniform(150000, 300000), 2),
120
+ "avg_response_ms": round(random.uniform(12, 45), 1), "model_accuracy": 0.9987,
121
+ "precision": 0.94, "recall": 0.89, "f1_score": 0.91, "roc_auc": 0.9823,
122
+ "hourly_stats": [{"hour": f"{i:02d}:00", "transactions": random.randint(400,1200), "fraud": random.randint(5,25)} for i in range(24)]}
123
+
124
+ @app.get("/shap/global")
125
+ async def shap_global():
126
+ feats = [f"V{i}" for i in range(1, 29)] + ["Amount"]
127
+ importance = sorted([{"feature": f, "importance": round(random.uniform(0.01, 0.35), 4)} for f in feats],
128
+ key=lambda x: x["importance"], reverse=True)
129
+ return {"feature_importance": importance[:15]}
data/eda_class_distribution.png ADDED
data/eda_correlations.png ADDED
data/evaluation_plots.png ADDED
data/shap_importance.png ADDED
data/shap_waterfall_fraud.png ADDED
docker-compose.yml ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ version: "3.8"
2
+
3
+ services:
4
+
5
+ # ── Zookeeper (required for Kafka) ────────────────────────────────────────
6
+ zookeeper:
7
+ image: confluentinc/cp-zookeeper:7.5.0
8
+ container_name: fraud_zookeeper
9
+ environment:
10
+ ZOOKEEPER_CLIENT_PORT: 2181
11
+ ZOOKEEPER_TICK_TIME: 2000
12
+ ports:
13
+ - "2181:2181"
14
+
15
+ # ── Kafka Broker ──────────────────────────────────────────────────────────
16
+ kafka:
17
+ image: confluentinc/cp-kafka:7.5.0
18
+ container_name: fraud_kafka
19
+ depends_on:
20
+ - zookeeper
21
+ ports:
22
+ - "9092:9092"
23
+ environment:
24
+ KAFKA_BROKER_ID: 1
25
+ KAFKA_ZOOKEEPER_CONNECT: zookeeper:2181
26
+ KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://localhost:9092
27
+ KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1
28
+ KAFKA_AUTO_CREATE_TOPICS_ENABLE: "true"
29
+
30
+ # ── FastAPI Scoring API ───────────────────────────────────────────────────
31
+ api:
32
+ build:
33
+ context: .
34
+ dockerfile: Dockerfile.api
35
+ container_name: fraud_api
36
+ ports:
37
+ - "8000:8000"
38
+ volumes:
39
+ - ./model:/app/model # mount trained model artifacts
40
+ environment:
41
+ MODEL_DIR: /app/model
42
+ depends_on:
43
+ - kafka
44
+ restart: unless-stopped
45
+
46
+ # ── React Frontend ────────────────────────────────────────────────────────
47
+ frontend:
48
+ build:
49
+ context: .
50
+ dockerfile: Dockerfile.frontend
51
+ args:
52
+ REACT_APP_API_URL: http://localhost:8000
53
+ container_name: fraud_frontend
54
+ ports:
55
+ - "3000:3000"
56
+ depends_on:
57
+ - api
58
+ restart: unless-stopped
59
+
60
+ networks:
61
+ default:
62
+ name: fraud_network
frontend/dashboard.html ADDED
@@ -0,0 +1,718 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!DOCTYPE html>
2
+ <html lang="en">
3
+ <head>
4
+ <meta charset="UTF-8">
5
+ <meta name="viewport" content="width=device-width, initial-scale=1.0">
6
+ <title>SENTINEL β€” Fraud Detection</title>
7
+ <link rel="preconnect" href="https://fonts.googleapis.com">
8
+ <link href="https://fonts.googleapis.com/css2?family=Share+Tech+Mono&family=Rajdhani:wght@400;500;600;700&family=Orbitron:wght@400;700;900&display=swap" rel="stylesheet">
9
+ <script src="https://cdnjs.cloudflare.com/ajax/libs/Chart.js/4.4.1/chart.umd.min.js"></script>
10
+ <style>
11
+ :root {
12
+ --bg: #04080f;
13
+ --bg2: #080f1a;
14
+ --bg3: #0c1824;
15
+ --border: #152840;
16
+ --cyan: #00d4ff;
17
+ --green: #00ff88;
18
+ --red: #ff2d55;
19
+ --orange: #ff9500;
20
+ --yellow: #ffd60a;
21
+ --txt: #dceeff;
22
+ --txt2: #6a9bbf;
23
+ --txt3: #2d5070;
24
+ --sidebar: 230px;
25
+ --mono: 'Share Tech Mono', monospace;
26
+ --body: 'Rajdhani', sans-serif;
27
+ --display: 'Orbitron', monospace;
28
+ }
29
+ *{margin:0;padding:0;box-sizing:border-box}
30
+ html,body{height:100%;overflow:hidden}
31
+ body{background:var(--bg);color:var(--txt);font-family:var(--body);font-size:14px;display:flex}
32
+
33
+ /* ── SIDEBAR ── */
34
+ .sidebar{
35
+ width:var(--sidebar);flex-shrink:0;
36
+ background:var(--bg2);border-right:1px solid var(--border);
37
+ display:flex;flex-direction:column;position:relative;z-index:10;
38
+ }
39
+ .sidebar::after{content:'';position:absolute;top:0;right:0;bottom:0;width:1px;background:linear-gradient(180deg,transparent,rgba(0,212,255,.3),transparent)}
40
+ .sb-logo{padding:20px 16px 14px;border-bottom:1px solid var(--border)}
41
+ .sb-logo-name{font-family:var(--display);font-size:16px;font-weight:900;color:var(--cyan);letter-spacing:3px;text-shadow:0 0 20px rgba(0,212,255,.4)}
42
+ .sb-logo-sub{font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px;margin-top:3px}
43
+ .sb-pill{display:inline-flex;align-items:center;gap:5px;background:rgba(0,255,136,.08);border:1px solid rgba(0,255,136,.25);border-radius:20px;padding:3px 9px;margin-top:8px;font-family:var(--mono);font-size:8px;color:var(--green)}
44
+ .sb-dot{width:5px;height:5px;background:var(--green);border-radius:50%;animation:pg 2s infinite}
45
+ @keyframes pg{0%,100%{opacity:1}50%{opacity:.3}}
46
+ .sb-nav{padding:14px 10px;flex:1}
47
+ .sb-sec{font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px;padding:0 8px;margin-bottom:6px;margin-top:4px}
48
+ .nav-btn{
49
+ display:flex;align-items:center;gap:10px;width:100%;
50
+ padding:9px 12px;border-radius:6px;border:1px solid transparent;
51
+ background:none;color:var(--txt2);cursor:pointer;
52
+ font-family:var(--body);font-size:13px;font-weight:500;letter-spacing:.3px;
53
+ transition:all .2s;margin-bottom:2px;text-align:left;position:relative;
54
+ }
55
+ .nav-btn:hover{background:var(--bg3);color:var(--txt);border-color:var(--border)}
56
+ .nav-btn.active{background:rgba(0,212,255,.07);color:var(--cyan);border-color:rgba(0,212,255,.2)}
57
+ .nav-btn.active::before{content:'';position:absolute;left:0;top:22%;bottom:22%;width:2px;background:var(--cyan);border-radius:2px;box-shadow:0 0 8px var(--cyan)}
58
+ .nav-icon{font-size:14px;width:18px;text-align:center;flex-shrink:0}
59
+ .sb-footer{padding:12px;border-top:1px solid var(--border)}
60
+ .sb-api{font-family:var(--mono);font-size:8px;color:var(--txt3);display:flex;align-items:center;gap:5px}
61
+
62
+ /* ── MAIN ── */
63
+ .main{flex:1;display:flex;flex-direction:column;overflow:hidden}
64
+ .topbar{
65
+ height:54px;background:var(--bg2);border-bottom:1px solid var(--border);
66
+ display:flex;align-items:center;justify-content:space-between;padding:0 22px;flex-shrink:0;
67
+ }
68
+ .topbar-title{font-family:var(--display);font-size:11px;font-weight:700;letter-spacing:2px}
69
+ .topbar-r{display:flex;align-items:center;gap:14px}
70
+ .live-badge{display:flex;align-items:center;gap:4px;background:rgba(255,45,85,.1);border:1px solid rgba(255,45,85,.35);border-radius:4px;padding:3px 9px;font-family:var(--mono);font-size:9px;color:var(--red);letter-spacing:1px}
71
+ .live-dot{width:5px;height:5px;background:var(--red);border-radius:50%;animation:bk 1s infinite}
72
+ @keyframes bk{0%,100%{opacity:1}50%{opacity:.15}}
73
+ .clock{font-family:var(--mono);font-size:11px;color:var(--cyan)}
74
+ .content{flex:1;overflow-y:auto;padding:20px}
75
+
76
+ /* ── SECTIONS ── */
77
+ .section{display:none;animation:fadeIn .3s ease}
78
+ .section.active{display:block}
79
+ @keyframes fadeIn{from{opacity:0;transform:translateY(8px)}to{opacity:1;transform:translateY(0)}}
80
+
81
+ /* ── STAT CARDS ── */
82
+ .stats-grid{display:grid;grid-template-columns:repeat(4,1fr);gap:14px;margin-bottom:20px}
83
+ .stat{background:var(--bg3);border:1px solid var(--border);border-radius:8px;padding:18px;position:relative;overflow:hidden;transition:border-color .2s}
84
+ .stat:hover{border-color:rgba(0,212,255,.3)}
85
+ .stat::before{content:'';position:absolute;top:0;left:0;right:0;height:2px}
86
+ .stat.c::before{background:var(--cyan)}.stat.g::before{background:var(--green)}
87
+ .stat.r::before{background:var(--red)}.stat.o::before{background:var(--orange)}
88
+ .stat-lbl{font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px;margin-bottom:8px}
89
+ .stat-val{font-family:var(--display);font-size:26px;font-weight:700;line-height:1;margin-bottom:4px}
90
+ .stat.c .stat-val{color:var(--cyan)}.stat.g .stat-val{color:var(--green)}
91
+ .stat.r .stat-val{color:var(--red)}.stat.o .stat-val{color:var(--orange)}
92
+ .stat-sub{font-family:var(--mono);font-size:9px;color:var(--txt3)}
93
+ .stat-ico{position:absolute;right:14px;top:14px;font-size:24px;opacity:.12}
94
+
95
+ /* ── CARDS ── */
96
+ .card{background:var(--bg3);border:1px solid var(--border);border-radius:8px;overflow:hidden;margin-bottom:16px}
97
+ .card-h{display:flex;align-items:center;justify-content:space-between;padding:12px 18px;border-bottom:1px solid var(--border)}
98
+ .card-t{font-family:var(--display);font-size:10px;font-weight:700;letter-spacing:2px}
99
+ .card-b{padding:18px}
100
+ .g2{display:grid;grid-template-columns:1fr 1fr;gap:16px}
101
+ .g3{display:grid;grid-template-columns:1.7fr 1fr;gap:16px}
102
+
103
+ /* ── TABLE ── */
104
+ .tbl{width:100%;border-collapse:collapse}
105
+ .tbl th{font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px;text-align:left;padding:8px 12px;border-bottom:1px solid var(--border);background:rgba(0,0,0,.2)}
106
+ .tbl td{padding:9px 12px;border-bottom:1px solid rgba(21,40,64,.5);font-size:11px;font-family:var(--mono)}
107
+ .tbl tr:hover td{background:rgba(255,255,255,.015)}
108
+ .badge{display:inline-flex;align-items:center;gap:3px;padding:2px 7px;border-radius:3px;font-size:8px;font-weight:bold;letter-spacing:1px;font-family:var(--mono)}
109
+ .b-fraud{background:rgba(255,45,85,.15);color:var(--red);border:1px solid rgba(255,45,85,.3)}
110
+ .b-legit{background:rgba(0,255,136,.08);color:var(--green);border:1px solid rgba(0,255,136,.2)}
111
+ .b-h{background:rgba(255,45,85,.15);color:var(--red);border:1px solid rgba(255,45,85,.3)}
112
+ .b-m{background:rgba(255,149,0,.12);color:var(--orange);border:1px solid rgba(255,149,0,.3)}
113
+ .b-l{background:rgba(0,255,136,.08);color:var(--green);border:1px solid rgba(0,255,136,.2)}
114
+
115
+ /* ── FORMS ── */
116
+ .form-lbl{display:block;font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px;margin-bottom:5px}
117
+ .form-inp{width:100%;background:var(--bg2);border:1px solid var(--border);border-radius:5px;padding:8px 12px;color:var(--txt);font-family:var(--mono);font-size:11px;outline:none;transition:border-color .2s}
118
+ .form-inp:focus{border-color:var(--cyan);box-shadow:0 0 0 2px rgba(0,212,255,.08)}
119
+ .fg{display:grid;grid-template-columns:repeat(4,1fr);gap:10px;margin-bottom:12px}
120
+ .btn{display:inline-flex;align-items:center;gap:6px;padding:8px 16px;border-radius:5px;border:1px solid;cursor:pointer;font-family:var(--display);font-size:10px;font-weight:700;letter-spacing:1px;transition:all .2s}
121
+ .btn-c{background:rgba(0,212,255,.08);border-color:var(--cyan);color:var(--cyan)}
122
+ .btn-c:hover{background:rgba(0,212,255,.18);box-shadow:0 0 16px rgba(0,212,255,.25)}
123
+ .btn-r{background:rgba(255,45,85,.08);border-color:var(--red);color:var(--red)}
124
+ .btn-r:hover{background:rgba(255,45,85,.18)}
125
+ .btn-g{background:rgba(0,255,136,.06);border-color:var(--green);color:var(--green)}
126
+
127
+ /* ── METRICS ── */
128
+ .mrow{display:flex;align-items:center;justify-content:space-between;padding:9px 0;border-bottom:1px solid rgba(21,40,64,.5)}
129
+ .mrow:last-child{border:none}
130
+ .mname{font-family:var(--mono);font-size:10px;color:var(--txt2)}
131
+ .mval{font-family:var(--display);font-size:13px;font-weight:700;color:var(--cyan)}
132
+ .ptrack{background:var(--border);border-radius:3px;height:5px;overflow:hidden;flex:1;margin:0 12px}
133
+ .pfill{height:100%;border-radius:3px;background:linear-gradient(90deg,var(--cyan),var(--green));transition:width 1.2s ease}
134
+
135
+ /* ── SHAP ── */
136
+ .shap-row{display:flex;align-items:center;gap:8px;padding:5px 0}
137
+ .shap-f{font-family:var(--mono);font-size:10px;color:var(--txt2);width:55px;flex-shrink:0}
138
+ .shap-bc{flex:1;height:18px;background:var(--bg2);border-radius:3px;overflow:hidden;position:relative}
139
+ .shap-pos{position:absolute;left:50%;top:0;bottom:0;background:rgba(255,45,85,.55);border-radius:0 3px 3px 0}
140
+ .shap-neg{position:absolute;right:50%;top:0;bottom:0;background:rgba(0,212,255,.55);border-radius:3px 0 0 3px}
141
+ .shap-v{font-family:var(--mono);font-size:9px;color:var(--txt3);width:55px;text-align:right;flex-shrink:0}
142
+
143
+ /* ── FEED ── */
144
+ .feed-item{
145
+ display:grid;grid-template-columns:85px 1fr 70px 85px 80px 65px;
146
+ align-items:center;gap:10px;padding:8px 14px;
147
+ border-bottom:1px solid rgba(21,40,64,.4);
148
+ font-family:var(--mono);font-size:10px;transition:background .15s;
149
+ animation:si .3s ease;
150
+ }
151
+ .feed-item:hover{background:rgba(255,255,255,.015)}
152
+ .feed-item.fraud-row{background:rgba(255,45,85,.03);border-left:2px solid var(--red)}
153
+ @keyframes si{from{opacity:0;transform:translateX(-6px)}to{opacity:1;transform:translateX(0)}}
154
+ .feed-hdr{
155
+ display:grid;grid-template-columns:85px 1fr 70px 85px 80px 65px;
156
+ gap:10px;padding:7px 14px;
157
+ background:rgba(0,0,0,.2);border-bottom:1px solid var(--border);
158
+ }
159
+ .fh{font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px}
160
+
161
+ /* ── RING ── */
162
+ .ring-wrap{position:relative;width:110px;height:110px;margin:0 auto}
163
+ .ring-wrap svg{transform:rotate(-90deg)}
164
+ .ring-txt{position:absolute;inset:0;display:flex;flex-direction:column;align-items:center;justify-content:center}
165
+ .ring-num{font-family:var(--display);font-size:20px;font-weight:900}
166
+ .ring-lbl{font-family:var(--mono);font-size:7px;color:var(--txt3);letter-spacing:1px}
167
+
168
+ /* ── CONFUSION MATRIX ── */
169
+ .cm-grid{display:grid;grid-template-columns:1fr 1fr;gap:10px;max-width:340px;margin:0 auto}
170
+ .cm-cell{text-align:center;padding:20px 10px;border-radius:7px;border:1px solid}
171
+ .cm-n{font-family:var(--display);font-size:28px;font-weight:700;margin-bottom:4px}
172
+ .cm-lbl{font-family:var(--mono);font-size:8px;letter-spacing:1px;margin-bottom:2px}
173
+ .cm-sub{font-family:var(--mono);font-size:8px}
174
+
175
+ /* ── SCROLLBAR ── */
176
+ ::-webkit-scrollbar{width:4px;height:4px}
177
+ ::-webkit-scrollbar-track{background:var(--bg)}
178
+ ::-webkit-scrollbar-thumb{background:var(--border);border-radius:2px}
179
+ ::-webkit-scrollbar-thumb:hover{background:var(--cyan)}
180
+
181
+ /* ── RESULT BOX ── */
182
+ .result-box{background:var(--bg2);border:1px solid var(--border);border-radius:7px;padding:16px;font-family:var(--mono);font-size:11px;min-height:120px}
183
+ .result-fraud{border-color:rgba(255,45,85,.4);background:rgba(255,45,85,.04)}
184
+ .result-legit{border-color:rgba(0,255,136,.25);background:rgba(0,255,136,.03)}
185
+
186
+ /* ── ALERT ── */
187
+ .alert-item{padding:9px 12px;border-radius:5px;font-family:var(--mono);font-size:9px;margin-bottom:7px;border-left:2px solid var(--red);background:rgba(255,45,85,.06);color:#ff6b87;animation:si .3s ease}
188
+
189
+ /* ── LOADING ── */
190
+ .spin{width:24px;height:24px;border:2px solid var(--border);border-top-color:var(--cyan);border-radius:50%;animation:sp .7s linear infinite;margin:0 auto}
191
+ @keyframes sp{to{transform:rotate(360deg)}}
192
+
193
+ /* SCANLINE */
194
+ body::after{content:'';position:fixed;inset:0;pointer-events:none;z-index:9999;background:repeating-linear-gradient(0deg,rgba(0,0,0,.03) 0px,rgba(0,0,0,.03) 1px,transparent 1px,transparent 2px);opacity:.4}
195
+ </style>
196
+ </head>
197
+ <body>
198
+
199
+ <!-- SIDEBAR -->
200
+ <div class="sidebar">
201
+ <div class="sb-logo">
202
+ <div class="sb-logo-name">SENTINEL</div>
203
+ <div class="sb-logo-sub">FRAUD DETECTION SYSTEM</div>
204
+ <div class="sb-pill"><span class="sb-dot"></span>SYSTEM ONLINE</div>
205
+ </div>
206
+ <div class="sb-nav">
207
+ <div class="sb-sec">NAVIGATION</div>
208
+ <button class="nav-btn active" onclick="showSection('overview',this)"><span class="nav-icon">⊞</span>Overview</button>
209
+ <button class="nav-btn" onclick="showSection('monitor',this)"><span class="nav-icon">β—‰</span>Live Monitor</button>
210
+ <button class="nav-btn" onclick="showSection('analyzer',this)"><span class="nav-icon">⬑</span>Analyzer</button>
211
+ <button class="nav-btn" onclick="showSection('performance',this)"><span class="nav-icon">β–²</span>Performance</button>
212
+ <button class="nav-btn" onclick="showSection('shap',this)"><span class="nav-icon">β—ˆ</span>SHAP Explain</button>
213
+ </div>
214
+ <div class="sb-footer">
215
+ <div class="sb-api" id="apiStatus">● CHECKING API...</div>
216
+ </div>
217
+ </div>
218
+
219
+ <!-- MAIN -->
220
+ <div class="main">
221
+ <div class="topbar">
222
+ <div class="topbar-title" id="topTitle">OVERVIEW / DASHBOARD</div>
223
+ <div class="topbar-r">
224
+ <div class="live-badge"><span class="live-dot"></span>LIVE</div>
225
+ <div class="clock" id="clock">--:--:--</div>
226
+ </div>
227
+ </div>
228
+ <div class="content">
229
+
230
+ <!-- ══ OVERVIEW ══ -->
231
+ <div class="section active" id="sec-overview">
232
+ <div class="stats-grid">
233
+ <div class="stat c"><div class="stat-ico">β—Ž</div><div class="stat-lbl">TOTAL TRANSACTIONS</div><div class="stat-val" id="ovTotal">β€”</div><div class="stat-sub">Last 24 hours</div></div>
234
+ <div class="stat r"><div class="stat-ico">⚠</div><div class="stat-lbl">FRAUD DETECTED</div><div class="stat-val" id="ovFraud">β€”</div><div class="stat-sub" id="ovFraudPct">β€” fraud rate</div></div>
235
+ <div class="stat g"><div class="stat-ico">β‚Ή</div><div class="stat-lbl">AMOUNT PROTECTED</div><div class="stat-val" id="ovSaved">β€”</div><div class="stat-sub">Blocked today</div></div>
236
+ <div class="stat o"><div class="stat-ico">⚑</div><div class="stat-lbl">AVG RESPONSE</div><div class="stat-val" id="ovMs">β€”</div><div class="stat-sub">Milliseconds</div></div>
237
+ </div>
238
+ <div class="g3">
239
+ <div class="card" style="margin:0">
240
+ <div class="card-h"><span class="card-t">TRANSACTION VOLUME β€” 24H</span></div>
241
+ <div class="card-b"><div style="height:200px"><canvas id="volChart"></canvas></div></div>
242
+ </div>
243
+ <div class="card" style="margin:0">
244
+ <div class="card-h"><span class="card-t">FRAUD SPLIT</span></div>
245
+ <div class="card-b"><div style="height:200px"><canvas id="donutChart"></canvas></div></div>
246
+ </div>
247
+ </div>
248
+ <div class="card">
249
+ <div class="card-h">
250
+ <span class="card-t">RECENT TRANSACTIONS</span>
251
+ <button class="btn btn-c" onclick="loadOverviewTxns()" style="padding:4px 12px;font-size:9px">↻ REFRESH</button>
252
+ </div>
253
+ <div id="ovTxnWrap"><div style="padding:24px;text-align:center"><div class="spin"></div></div></div>
254
+ </div>
255
+ </div>
256
+
257
+ <!-- ══ LIVE MONITOR ══ -->
258
+ <div class="section" id="sec-monitor">
259
+ <div class="stats-grid">
260
+ <div class="stat c"><div class="stat-lbl">PROCESSED (SESSION)</div><div class="stat-val" id="mTotal">0</div><div class="stat-sub">Auto-refresh 5s</div></div>
261
+ <div class="stat r"><div class="stat-lbl">FRAUDS CAUGHT</div><div class="stat-val" id="mFraud">0</div><div class="stat-sub" id="mFraudPct">0.00% rate</div></div>
262
+ <div class="stat g"><div class="stat-lbl">LEGIT PASSED</div><div class="stat-val" id="mLegit">0</div><div class="stat-sub">Safe transactions</div></div>
263
+ <div class="stat o"><div class="stat-lbl">HIGH RISK ALERTS</div><div class="stat-val" id="mHigh">0</div><div class="stat-sub">Needs review</div></div>
264
+ </div>
265
+ <div class="g3">
266
+ <div class="card" style="margin:0">
267
+ <div class="card-h">
268
+ <span class="card-t">LIVE FEED</span>
269
+ <button class="btn btn-c" id="pauseBtn" onclick="togglePause()" style="padding:4px 12px;font-size:9px">⏸ PAUSE</button>
270
+ </div>
271
+ <div class="feed-hdr">
272
+ <span class="fh">TXN ID</span><span class="fh">MERCHANT</span><span class="fh">AMOUNT</span>
273
+ <span class="fh">PROB</span><span class="fh">STATUS</span><span class="fh">TIME</span>
274
+ </div>
275
+ <div id="liveFeed" style="max-height:350px;overflow-y:auto"></div>
276
+ </div>
277
+ <div style="display:flex;flex-direction:column;gap:14px">
278
+ <div class="card" style="margin:0">
279
+ <div class="card-h"><span class="card-t">SESSION RINGS</span></div>
280
+ <div class="card-b" style="display:flex;justify-content:space-around">
281
+ <div>
282
+ <div class="ring-wrap">
283
+ <svg width="110" height="110" viewBox="0 0 110 110">
284
+ <circle cx="55" cy="55" r="46" fill="none" stroke="#152840" stroke-width="7"/>
285
+ <circle id="fraudRing" cx="55" cy="55" r="46" fill="none" stroke="#ff2d55" stroke-width="7" stroke-dasharray="289" stroke-dashoffset="289" stroke-linecap="round"/>
286
+ </svg>
287
+ <div class="ring-txt"><span class="ring-num" style="color:var(--red)" id="rFraud">0%</span><span class="ring-lbl">FRAUD</span></div>
288
+ </div>
289
+ </div>
290
+ <div>
291
+ <div class="ring-wrap">
292
+ <svg width="110" height="110" viewBox="0 0 110 110">
293
+ <circle cx="55" cy="55" r="46" fill="none" stroke="#152840" stroke-width="7"/>
294
+ <circle id="legitRing" cx="55" cy="55" r="46" fill="none" stroke="#00d4ff" stroke-width="7" stroke-dasharray="289" stroke-dashoffset="0" stroke-linecap="round"/>
295
+ </svg>
296
+ <div class="ring-txt"><span class="ring-num" style="color:var(--cyan)" id="rLegit">100%</span><span class="ring-lbl">LEGIT</span></div>
297
+ </div>
298
+ </div>
299
+ </div>
300
+ </div>
301
+ <div class="card" style="margin:0;flex:1">
302
+ <div class="card-h"><span class="card-t">⚠ ALERTS</span></div>
303
+ <div id="alertPanel" style="padding:10px;max-height:200px;overflow-y:auto">
304
+ <div style="font-family:var(--mono);font-size:9px;color:var(--txt3);text-align:center;padding:12px">No alerts yet...</div>
305
+ </div>
306
+ </div>
307
+ </div>
308
+ </div>
309
+ <div class="card">
310
+ <div class="card-h"><span class="card-t">FRAUD PROBABILITY TREND</span></div>
311
+ <div class="card-b"><div style="height:150px"><canvas id="trendChart"></canvas></div></div>
312
+ </div>
313
+ </div>
314
+
315
+ <!-- ══ ANALYZER ══ -->
316
+ <div class="section" id="sec-analyzer">
317
+ <div class="g2" style="align-items:start">
318
+ <div class="card" style="margin:0">
319
+ <div class="card-h">
320
+ <span class="card-t">TRANSACTION INPUT</span>
321
+ <div style="display:flex;gap:8px">
322
+ <button class="btn btn-c" onclick="fillRandom()" style="padding:4px 10px;font-size:9px">βš„ RANDOM</button>
323
+ <button class="btn btn-r" onclick="clearAnalyzer()" style="padding:4px 10px;font-size:9px">βœ• CLEAR</button>
324
+ </div>
325
+ </div>
326
+ <div class="card-b">
327
+ <div style="margin-bottom:12px">
328
+ <label class="form-lbl">AMOUNT (β‚Ή)</label>
329
+ <input class="form-inp" type="number" id="inpAmount" value="250" step="0.01" style="max-width:200px">
330
+ </div>
331
+ <div class="form-lbl" style="margin-bottom:8px">PCA FEATURES V1 – V28</div>
332
+ <div class="fg" id="vInputs"></div>
333
+ <div style="display:flex;gap:10px;margin-top:4px">
334
+ <button class="btn btn-c" onclick="runAnalyze()" id="analyzeBtn">⬑ ANALYZE</button>
335
+ <button class="btn btn-g" onclick="runBatch()" style="padding:8px 14px">β–Ά BATCH Γ—10</button>
336
+ </div>
337
+ </div>
338
+ </div>
339
+ <div style="display:flex;flex-direction:column;gap:14px">
340
+ <div class="card" style="margin:0">
341
+ <div class="card-h"><span class="card-t">GAUGE</span></div>
342
+ <div class="card-b" style="text-align:center">
343
+ <canvas id="gaugeChart" width="200" height="130" style="max-width:200px"></canvas>
344
+ <div id="gaugeLbl" style="font-family:var(--display);font-size:26px;font-weight:900;color:var(--txt3);margin-top:-28px">β€”</div>
345
+ <div style="font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px">FRAUD PROBABILITY</div>
346
+ </div>
347
+ </div>
348
+ <div class="card" style="margin:0">
349
+ <div class="card-h"><span class="card-t">RESULT</span></div>
350
+ <div class="card-b" id="analyzeResult">
351
+ <div style="font-family:var(--mono);font-size:10px;color:var(--txt3);text-align:center;padding:16px">Run analysis to see result...</div>
352
+ </div>
353
+ </div>
354
+ <div class="card" style="margin:0">
355
+ <div class="card-h"><span class="card-t">BATCH RESULT</span></div>
356
+ <div class="card-b" id="batchResult">
357
+ <div style="font-family:var(--mono);font-size:10px;color:var(--txt3);text-align:center;padding:8px">Click Batch Γ—10 to run</div>
358
+ </div>
359
+ </div>
360
+ </div>
361
+ </div>
362
+ </div>
363
+
364
+ <!-- ══ PERFORMANCE ══ -->
365
+ <div class="section" id="sec-performance">
366
+ <div class="stats-grid" style="grid-template-columns:repeat(5,1fr)">
367
+ <div class="stat c"><div class="stat-lbl">ROC-AUC</div><div class="stat-val" id="pAuc">β€”</div><div class="stat-sub">Area Under Curve</div></div>
368
+ <div class="stat g"><div class="stat-lbl">ACCURACY</div><div class="stat-val" id="pAcc">β€”</div><div class="stat-sub">Overall correct</div></div>
369
+ <div class="stat o"><div class="stat-lbl">PRECISION</div><div class="stat-val" id="pPrec">β€”</div><div class="stat-sub">Of flagged</div></div>
370
+ <div class="stat r"><div class="stat-lbl">RECALL</div><div class="stat-val" id="pRec">β€”</div><div class="stat-sub">Fraud caught</div></div>
371
+ <div class="stat c" style="border-top-color:var(--yellow)"><div class="stat-lbl">F1 SCORE</div><div class="stat-val" id="pF1" style="color:var(--yellow)">β€”</div><div class="stat-sub">Harmonic mean</div></div>
372
+ </div>
373
+ <div class="g2">
374
+ <div class="card" style="margin:0">
375
+ <div class="card-h"><span class="card-t">METRIC BARS</span></div>
376
+ <div class="card-b" id="metricBars"></div>
377
+ </div>
378
+ <div class="card" style="margin:0">
379
+ <div class="card-h"><span class="card-t">CONFUSION MATRIX</span></div>
380
+ <div class="card-b">
381
+ <div class="cm-grid">
382
+ <div class="cm-cell" style="background:rgba(0,255,136,.06);border-color:rgba(0,255,136,.2)"><div class="cm-n" style="color:var(--green)" id="cmTN">β€”</div><div class="cm-lbl" style="color:var(--txt3)">TRUE NEGATIVE</div><div class="cm-sub" style="color:var(--green)">Correct Legit</div></div>
383
+ <div class="cm-cell" style="background:rgba(255,149,0,.06);border-color:rgba(255,149,0,.2)"><div class="cm-n" style="color:var(--orange)" id="cmFP">β€”</div><div class="cm-lbl" style="color:var(--txt3)">FALSE POSITIVE</div><div class="cm-sub" style="color:var(--orange)">False Alarm</div></div>
384
+ <div class="cm-cell" style="background:rgba(255,45,85,.06);border-color:rgba(255,45,85,.2)"><div class="cm-n" style="color:var(--red)" id="cmFN">β€”</div><div class="cm-lbl" style="color:var(--txt3)">FALSE NEGATIVE</div><div class="cm-sub" style="color:var(--red)">Missed Fraud</div></div>
385
+ <div class="cm-cell" style="background:rgba(0,212,255,.06);border-color:rgba(0,212,255,.2)"><div class="cm-n" style="color:var(--cyan)" id="cmTP">β€”</div><div class="cm-lbl" style="color:var(--txt3)">TRUE POSITIVE</div><div class="cm-sub" style="color:var(--cyan)">Caught Fraud</div></div>
386
+ </div>
387
+ </div>
388
+ </div>
389
+ </div>
390
+ <div class="g2">
391
+ <div class="card" style="margin:0">
392
+ <div class="card-h"><span class="card-t">HOURLY FRAUD TREND</span></div>
393
+ <div class="card-b"><div style="height:180px"><canvas id="perfChart"></canvas></div></div>
394
+ </div>
395
+ <div class="card" style="margin:0">
396
+ <div class="card-h"><span class="card-t">THRESHOLD TUNER</span></div>
397
+ <div class="card-b">
398
+ <label class="form-lbl">THRESHOLD: <span id="threshVal" style="color:var(--cyan)">0.50</span></label>
399
+ <input type="range" id="threshSlider" min="0.1" max="0.9" step="0.05" value="0.5"
400
+ style="width:100%;accent-color:var(--cyan);margin:10px 0 16px"
401
+ oninput="updateThresh(this.value)">
402
+ <div id="threshStats"></div>
403
+ </div>
404
+ </div>
405
+ </div>
406
+ </div>
407
+
408
+ <!-- ══ SHAP ══ -->
409
+ <div class="section" id="sec-shap">
410
+ <div style="background:rgba(0,212,255,.05);border:1px solid rgba(0,212,255,.2);border-radius:6px;padding:10px 14px;font-family:var(--mono);font-size:10px;color:var(--cyan);margin-bottom:16px">
411
+ β—ˆ SHAP shows which features push prediction toward FRAUD (red) or LEGITIMATE (blue). Longer bar = stronger influence.
412
+ </div>
413
+ <div class="g3">
414
+ <div class="card" style="margin:0">
415
+ <div class="card-h"><span class="card-t">GLOBAL FEATURE IMPORTANCE</span></div>
416
+ <div class="card-b"><div style="height:320px"><canvas id="shapChart"></canvas></div></div>
417
+ </div>
418
+ <div style="display:flex;flex-direction:column;gap:14px">
419
+ <div class="card" style="margin:0">
420
+ <div class="card-h"><span class="card-t">LEGEND</span></div>
421
+ <div class="card-b">
422
+ <div style="display:flex;align-items:center;gap:8px;margin-bottom:8px"><div style="width:32px;height:8px;background:var(--red);border-radius:2px"></div><span style="font-family:var(--mono);font-size:9px;color:var(--txt2)">Pushes β†’ FRAUD</span></div>
423
+ <div style="display:flex;align-items:center;gap:8px;margin-bottom:14px"><div style="width:32px;height:8px;background:var(--cyan);border-radius:2px"></div><span style="font-family:var(--mono);font-size:9px;color:var(--txt2)">Pushes β†’ LEGIT</span></div>
424
+ <div style="font-family:var(--mono);font-size:9px;color:var(--txt3);line-height:1.8">V14 and V4 are typically the strongest fraud signals in the Credit Card dataset.</div>
425
+ </div>
426
+ </div>
427
+ <div class="card" style="margin:0">
428
+ <div class="card-h"><span class="card-t">TOP INDICATORS</span></div>
429
+ <div class="card-b" id="topFeatures"></div>
430
+ </div>
431
+ </div>
432
+ </div>
433
+ <div class="card">
434
+ <div class="card-h">
435
+ <span class="card-t">LOCAL EXPLANATION</span>
436
+ <button class="btn btn-c" onclick="explainRandom()" style="padding:4px 12px;font-size:9px">βš„ EXPLAIN RANDOM</button>
437
+ </div>
438
+ <div class="card-b" id="localShap">
439
+ <div style="font-family:var(--mono);font-size:10px;color:var(--txt3);text-align:center;padding:16px">Click "EXPLAIN RANDOM" to see a local explanation.</div>
440
+ </div>
441
+ </div>
442
+ </div>
443
+
444
+ </div><!-- end .content -->
445
+ </div><!-- end .main -->
446
+
447
+ <script>
448
+ const API = 'http://localhost:8000';
449
+
450
+ // ── utils ──────────────────────────────────────────────────
451
+ async function get(path){ try{ const r=await fetch(API+path); return r.ok?await r.json():null; }catch(e){ return null; } }
452
+ async function post(path,body){ try{ const r=await fetch(API+path,{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify(body)}); return r.ok?await r.json():null; }catch(e){ return null; } }
453
+ function fc(n){ if(n>=1e6) return 'β‚Ή'+(n/1e6).toFixed(1)+'M'; if(n>=1e3) return 'β‚Ή'+(n/1e3).toFixed(1)+'K'; return 'β‚Ή'+n.toFixed(0); }
454
+ function fp(n){ return (n*100).toFixed(2)+'%'; }
455
+ function bb(isFraud){ return isFraud?`<span class="badge b-fraud">⚠ FRAUD</span>`:`<span class="badge b-legit">βœ“ LEGIT</span>`; }
456
+ function rb(lvl){ const m={HIGH:'b-h',MEDIUM:'b-m',LOW:'b-l'}; return `<span class="badge ${m[lvl]||'b-l'}">${lvl}</span>`; }
457
+ function pb(prob){ const p=(prob*100).toFixed(0); const c=prob>.7?'#ff2d55':prob>.4?'#ff9500':'#00ff88'; return `<span style="display:inline-block;width:60px;height:4px;background:#152840;border-radius:2px;vertical-align:middle;margin-right:4px;overflow:hidden"><span style="display:block;height:100%;width:${p}%;background:${c};border-radius:2px"></span></span>${p}%`; }
458
+ function ta(ts){ const d=Math.floor((Date.now()-new Date(ts))/1000); if(d<60) return d+'s'; if(d<3600) return Math.floor(d/60)+'m'; return Math.floor(d/3600)+'h'; }
459
+
460
+ // ── navigation ─────────────────────────────────────────────
461
+ const titles = { overview:'OVERVIEW / DASHBOARD', monitor:'LIVE MONITOR', analyzer:'TRANSACTION ANALYZER', performance:'MODEL PERFORMANCE', shap:'SHAP EXPLAINABILITY' };
462
+ function showSection(id, btn){
463
+ document.querySelectorAll('.section').forEach(s=>s.classList.remove('active'));
464
+ document.querySelectorAll('.nav-btn').forEach(b=>b.classList.remove('active'));
465
+ document.getElementById('sec-'+id).classList.add('active');
466
+ btn.classList.add('active');
467
+ document.getElementById('topTitle').textContent = titles[id];
468
+ if(id==='overview') loadOverview();
469
+ if(id==='monitor') startMonitor();
470
+ if(id==='performance') loadPerformance();
471
+ if(id==='shap') loadShap();
472
+ if(id==='analyzer') buildVInputs();
473
+ }
474
+
475
+ // ── clock ──────────────────────────────────────────────────
476
+ setInterval(()=>{ document.getElementById('clock').textContent=new Date().toLocaleTimeString('en-IN',{hour12:false})+' IST'; },1000);
477
+
478
+ // ── API health ─────────────────────────────────────────────
479
+ async function checkHealth(){
480
+ const d=await get('/health');
481
+ const el=document.getElementById('apiStatus');
482
+ el.innerHTML=d&&d.status==='online'?'<span style="color:var(--green)">● API ONLINE</span>':'<span style="color:var(--red)">● API OFFLINE</span>';
483
+ }
484
+ checkHealth(); setInterval(checkHealth,10000);
485
+
486
+ // ── chart helpers ──────────────────────────────────────────
487
+ function lineChart(id, labels, datasets){
488
+ const ctx=document.getElementById(id); if(!ctx) return null;
489
+ if(ctx._c) ctx._c.destroy();
490
+ ctx._c=new Chart(ctx,{ type:'line', data:{labels,datasets}, options:{ responsive:true, maintainAspectRatio:false, plugins:{legend:{labels:{color:'#6a9bbf',font:{family:'Share Tech Mono',size:9}}}}, scales:{ x:{grid:{color:'#15284055'},ticks:{color:'#6a9bbf',font:{family:'Share Tech Mono',size:8}}}, y:{grid:{color:'#15284055'},ticks:{color:'#6a9bbf',font:{family:'Share Tech Mono',size:8}}} } } });
491
+ return ctx._c;
492
+ }
493
+ function barChart(id, labels, datasets, opts={}){
494
+ const ctx=document.getElementById(id); if(!ctx) return null;
495
+ if(ctx._c) ctx._c.destroy();
496
+ ctx._c=new Chart(ctx,{ type:'bar', data:{labels,datasets}, options:{ responsive:true, maintainAspectRatio:false, plugins:{legend:{display:false}}, scales:{ x:{grid:{color:'#15284055'},ticks:{color:'#6a9bbf',font:{family:'Share Tech Mono',size:8}}}, y:{grid:{color:'#15284055'},ticks:{color:'#6a9bbf',font:{family:'Share Tech Mono',size:8}}} }, ...opts } });
497
+ return ctx._c;
498
+ }
499
+ function doughnut(id, labels, data, colors){
500
+ const ctx=document.getElementById(id); if(!ctx) return;
501
+ if(ctx._c) ctx._c.destroy();
502
+ ctx._c=new Chart(ctx,{ type:'doughnut', data:{labels,datasets:[{data,backgroundColor:colors,borderWidth:0}]}, options:{ responsive:true, maintainAspectRatio:false, cutout:'70%', plugins:{legend:{position:'bottom',labels:{color:'#6a9bbf',font:{family:'Share Tech Mono',size:9},padding:10}}} } });
503
+ }
504
+
505
+ // ══════════════════════════════════════════════════════════
506
+ // OVERVIEW
507
+ // ══════════════════════════════════════════════════════════
508
+ async function loadOverview(){
509
+ const d=await get('/stats/summary'); if(!d) return;
510
+ document.getElementById('ovTotal').textContent=d.total_transactions.toLocaleString();
511
+ document.getElementById('ovFraud').textContent=d.fraud_detected.toLocaleString();
512
+ document.getElementById('ovFraudPct').textContent=fp(d.fraud_rate)+' fraud rate';
513
+ document.getElementById('ovSaved').textContent=fc(d.amount_saved);
514
+ document.getElementById('ovMs').textContent=d.avg_response_ms+'ms';
515
+ const hrs=d.hourly_stats.map(h=>h.hour.slice(0,5));
516
+ lineChart('volChart',hrs,[
517
+ {label:'Transactions',data:d.hourly_stats.map(h=>h.transactions),borderColor:'#00d4ff',backgroundColor:'rgba(0,212,255,.07)',tension:.4,fill:true,pointRadius:0},
518
+ {label:'Fraud',data:d.hourly_stats.map(h=>h.fraud),borderColor:'#ff2d55',backgroundColor:'rgba(255,45,85,.05)',tension:.4,fill:true,pointRadius:0}
519
+ ]);
520
+ doughnut('donutChart',['Legitimate','Fraud'],[d.total_transactions-d.fraud_detected,d.fraud_detected],['#00d4ff','#ff2d55']);
521
+ loadOverviewTxns();
522
+ }
523
+
524
+ async function loadOverviewTxns(){
525
+ document.getElementById('ovTxnWrap').innerHTML='<div style="padding:20px;text-align:center"><div class="spin"></div></div>';
526
+ const d=await get('/stream/live');
527
+ if(!d){ document.getElementById('ovTxnWrap').innerHTML='<div style="padding:16px;font-family:var(--mono);font-size:10px;color:var(--red);text-align:center">API offline β€” run: uvicorn api.main:app --port 8000</div>'; return; }
528
+ const rows=d.transactions.map(t=>`<tr><td style="color:var(--cyan)">${t.id}</td><td>${t.merchant}</td><td>${t.location}</td><td>β‚Ή${t.amount.toLocaleString()}</td><td>${pb(t.fraud_probability)}</td><td>${rb(t.risk_level)}</td><td>${bb(t.is_fraud)}</td><td style="color:var(--txt3)">${ta(t.timestamp)}</td></tr>`).join('');
529
+ document.getElementById('ovTxnWrap').innerHTML=`<div style="overflow-x:auto"><table class="tbl"><thead><tr><th>TXN ID</th><th>MERCHANT</th><th>LOCATION</th><th>AMOUNT</th><th>PROB</th><th>RISK</th><th>STATUS</th><th>TIME</th></tr></thead><tbody>${rows}</tbody></table></div>`;
530
+ }
531
+
532
+ // ══════════════════════════════════════════════════════════
533
+ // MONITOR
534
+ // ═══════════════════════════════════════════════════���══════
535
+ let paused=false, mTotal=0, mFraud=0, mHigh=0, monitorStarted=false;
536
+ const trendLabels=[], trendData=[];
537
+ let trendChart=null;
538
+
539
+ function togglePause(){ paused=!paused; document.getElementById('pauseBtn').textContent=paused?'β–Ά RESUME':'⏸ PAUSE'; }
540
+
541
+ function startMonitor(){ if(!monitorStarted){ monitorStarted=true; fetchMonitor(); setInterval(fetchMonitor,5000); } }
542
+
543
+ async function fetchMonitor(){
544
+ if(paused) return;
545
+ const d=await get('/stream/live'); if(!d) return;
546
+ const txns=d.transactions.slice(0,8);
547
+ mTotal+=txns.length; const nf=txns.filter(t=>t.is_fraud).length; const nh=txns.filter(t=>t.risk_level==='HIGH').length;
548
+ mFraud+=nf; mHigh+=nh;
549
+ document.getElementById('mTotal').textContent=mTotal.toLocaleString();
550
+ document.getElementById('mFraud').textContent=mFraud.toLocaleString();
551
+ document.getElementById('mLegit').textContent=(mTotal-mFraud).toLocaleString();
552
+ document.getElementById('mHigh').textContent=mHigh.toLocaleString();
553
+ const fp=mTotal>0?(mFraud/mTotal*100):0;
554
+ document.getElementById('mFraudPct').textContent=fp.toFixed(2)+'% rate';
555
+ const c=289; document.getElementById('fraudRing').style.strokeDashoffset=c-(fp/100)*c;
556
+ document.getElementById('legitRing').style.strokeDashoffset=c-((100-fp)/100)*c;
557
+ document.getElementById('rFraud').textContent=fp.toFixed(1)+'%';
558
+ document.getElementById('rLegit').textContent=(100-fp).toFixed(1)+'%';
559
+ const now=new Date().toLocaleTimeString('en',{hour12:false});
560
+ trendLabels.push(now); trendData.push(+(txns.reduce((s,t)=>s+t.fraud_probability,0)/txns.length*100).toFixed(1));
561
+ if(trendLabels.length>20){ trendLabels.shift(); trendData.shift(); }
562
+ if(!trendChart){
563
+ trendChart=lineChart('trendChart',trendLabels,[{label:'Avg Fraud %',data:trendData,borderColor:'#ff2d55',backgroundColor:'rgba(255,45,85,.07)',tension:.4,fill:true,pointRadius:2,pointBackgroundColor:'#ff2d55'}]);
564
+ } else { trendChart.data.labels=trendLabels; trendChart.data.datasets[0].data=trendData; trendChart.update('none'); }
565
+ document.getElementById('liveFeed').innerHTML=txns.map(t=>`<div class="feed-item ${t.is_fraud?'fraud-row':''}"><span style="color:var(--cyan)">${t.id}</span><span>${t.merchant}<br><span style="color:var(--txt3);font-size:8px">${t.location}</span></span><span>β‚Ή${t.amount.toLocaleString()}</span><span>${pb(t.fraud_probability)}</span><span>${bb(t.is_fraud)}</span><span style="color:var(--txt3)">${ta(t.timestamp)}</span></div>`).join('');
566
+ txns.filter(t=>t.is_fraud).forEach(t=>{ const p=document.getElementById('alertPanel'); const noAlerts=p.querySelector('[data-no-alerts]'); if(noAlerts) noAlerts.remove(); const el=document.createElement('div'); el.className='alert-item'; el.innerHTML=`⚠ <strong>${t.id}</strong> β€” β‚Ή${t.amount.toLocaleString()} at ${t.merchant} β€” ${(t.fraud_probability*100).toFixed(1)}%`; p.prepend(el); if(p.children.length>7) p.lastChild.remove(); });
567
+ }
568
+
569
+ // ══════════════════════════════════════════════════════════
570
+ // ANALYZER
571
+ // ══════════════════════════════════════════════════════════
572
+ function buildVInputs(){
573
+ if(document.getElementById('vInputs').children.length>0) return;
574
+ let h='';
575
+ for(let i=1;i<=28;i++) h+=`<div><label class="form-lbl">V${i}</label><input class="form-inp" type="number" id="V${i}" value="0" step="0.001" style="padding:5px 8px;font-size:10px"></div>`;
576
+ document.getElementById('vInputs').innerHTML=h;
577
+ }
578
+
579
+ function fillRandom(){
580
+ buildVInputs();
581
+ const fraud=Math.random()<.3;
582
+ document.getElementById('inpAmount').value=fraud?(Math.random()*400+10).toFixed(2):(Math.random()*3000+100).toFixed(2);
583
+ for(let i=1;i<=28;i++){ const el=document.getElementById('V'+i); if(el) el.value=(Math.random()*6-3).toFixed(4); }
584
+ if(fraud){ document.getElementById('V14').value=(-7+Math.random()*1.5).toFixed(4); document.getElementById('V4').value=(4+Math.random()*2).toFixed(4); }
585
+ }
586
+
587
+ function clearAnalyzer(){
588
+ document.getElementById('inpAmount').value='0';
589
+ for(let i=1;i<=28;i++){ const el=document.getElementById('V'+i); if(el) el.value='0'; }
590
+ document.getElementById('analyzeResult').innerHTML='<div style="font-family:var(--mono);font-size:10px;color:var(--txt3);text-align:center;padding:16px">Run analysis to see result...</div>';
591
+ updateGauge(null);
592
+ }
593
+
594
+ function updateGauge(prob){
595
+ const ctx=document.getElementById('gaugeChart'); if(!ctx) return;
596
+ if(ctx._c) ctx._c.destroy();
597
+ const p=prob===null?0:prob;
598
+ const color=p>.7?'#ff2d55':p>.4?'#ff9500':'#00ff88';
599
+ ctx._c=new Chart(ctx,{ type:'doughnut', data:{datasets:[{data:[p*100,100-p*100],backgroundColor:[color,'#152840'],borderWidth:0}]}, options:{ responsive:false, circumference:180, rotation:-90, cutout:'74%', plugins:{legend:{display:false},tooltip:{enabled:false}} } });
600
+ const lbl=document.getElementById('gaugeLbl');
601
+ lbl.textContent=prob===null?'β€”':(p*100).toFixed(1)+'%';
602
+ lbl.style.color=prob===null?'var(--txt3)':color;
603
+ }
604
+ updateGauge(null);
605
+
606
+ async function runAnalyze(){
607
+ const btn=document.getElementById('analyzeBtn'); btn.textContent='⟳...'; btn.disabled=true;
608
+ const payload={Amount:parseFloat(document.getElementById('inpAmount').value)||0};
609
+ for(let i=1;i<=28;i++) payload['V'+i]=parseFloat(document.getElementById('V'+i)?.value)||0;
610
+ const r=await post('/predict',payload);
611
+ btn.textContent='⬑ ANALYZE'; btn.disabled=false;
612
+ if(!r){ document.getElementById('analyzeResult').innerHTML='<div style="color:var(--red);font-family:var(--mono);font-size:10px">API not responding.</div>'; return; }
613
+ updateGauge(r.fraud_probability);
614
+ const color=r.is_fraud?'var(--red)':'var(--green)';
615
+ document.getElementById('analyzeResult').innerHTML=`
616
+ <div style="display:flex;align-items:center;gap:12px;margin-bottom:14px">
617
+ <div style="font-size:36px">${r.is_fraud?'⚠':'βœ“'}</div>
618
+ <div><div style="font-family:var(--display);font-size:15px;font-weight:900;color:${color}">${r.is_fraud?'FRAUD DETECTED':'TRANSACTION CLEAR'}</div>
619
+ <div style="font-family:var(--mono);font-size:9px;color:var(--txt3)">${(r.fraud_probability*100).toFixed(2)}% confidence</div></div>
620
+ </div>
621
+ <div class="mrow"><span class="mname">Amount</span><span class="mval">β‚Ή${(r.amount||payload.Amount).toLocaleString()}</span></div>
622
+ <div class="mrow"><span class="mname">Fraud Prob</span><span class="mval" style="color:${color}">${(r.fraud_probability*100).toFixed(2)}%</span></div>
623
+ <div class="mrow"><span class="mname">Risk Level</span><span>${rb(r.risk_level)}</span></div>
624
+ <div class="mrow"><span class="mname">Verdict</span><span>${bb(r.is_fraud)}</span></div>
625
+ ${r.demo_mode?'<div style="margin-top:8px;font-family:var(--mono);font-size:8px;color:var(--txt3)">⚠ Demo mode β€” model.pkl not loaded</div>':''}`;
626
+ }
627
+
628
+ async function runBatch(){
629
+ document.getElementById('batchResult').innerHTML='<div style="text-align:center;padding:12px"><div class="spin"></div></div>';
630
+ const transactions=Array.from({length:10},()=>{ const o={Amount:Math.random()*3000+10}; for(let i=1;i<=28;i++) o['V'+i]=+(Math.random()*6-3).toFixed(4); return o; });
631
+ const r=await post('/predict/batch',{transactions});
632
+ if(!r){ document.getElementById('batchResult').innerHTML='<div style="color:var(--red);font-family:var(--mono);font-size:10px">API offline</div>'; return; }
633
+ document.getElementById('batchResult').innerHTML=`
634
+ <div style="display:grid;grid-template-columns:1fr 1fr 1fr;gap:8px;margin-bottom:10px">
635
+ <div style="text-align:center;padding:8px;background:var(--bg2);border-radius:5px;border:1px solid var(--border)"><div style="font-family:var(--display);font-size:18px;color:var(--cyan)">${r.total}</div><div style="font-family:var(--mono);font-size:8px;color:var(--txt3)">TOTAL</div></div>
636
+ <div style="text-align:center;padding:8px;background:var(--bg2);border-radius:5px;border:1px solid rgba(255,45,85,.3)"><div style="font-family:var(--display);font-size:18px;color:var(--red)">${r.fraud_count}</div><div style="font-family:var(--mono);font-size:8px;color:var(--txt3)">FRAUD</div></div>
637
+ <div style="text-align:center;padding:8px;background:var(--bg2);border-radius:5px;border:1px solid rgba(0,255,136,.2)"><div style="font-family:var(--display);font-size:18px;color:var(--green)">${fp(r.fraud_rate)}</div><div style="font-family:var(--mono);font-size:8px;color:var(--txt3)">RATE</div></div>
638
+ </div>
639
+ <div style="font-family:var(--mono);font-size:9px">${r.results.map((x,i)=>`<span style="display:inline-block;margin:2px;padding:2px 6px;border-radius:3px;background:${x.is_fraud?'rgba(255,45,85,.15)':'rgba(0,255,136,.07)'}">T${i+1}:${(x.fraud_probability*100).toFixed(0)}%</span>`).join('')}</div>`;
640
+ }
641
+
642
+ // ══════════════════════════════════════════════════════════
643
+ // PERFORMANCE
644
+ // ══════════════════════════════════════════════════════════
645
+ async function loadPerformance(){
646
+ const d=await get('/stats/summary'); if(!d) return;
647
+ document.getElementById('pAuc').textContent=d.roc_auc.toFixed(4);
648
+ document.getElementById('pAcc').textContent=d.model_accuracy.toFixed(4);
649
+ document.getElementById('pPrec').textContent=d.precision.toFixed(3);
650
+ document.getElementById('pRec').textContent=d.recall.toFixed(3);
651
+ document.getElementById('pF1').textContent=d.f1_score.toFixed(3);
652
+ const metrics=[{n:'ROC-AUC',v:d.roc_auc,c:'#00d4ff'},{n:'Accuracy',v:d.model_accuracy,c:'#00ff88'},{n:'Precision',v:d.precision,c:'#ff9500'},{n:'Recall',v:d.recall,c:'#ff2d55'},{n:'F1 Score',v:d.f1_score,c:'#ffd60a'}];
653
+ document.getElementById('metricBars').innerHTML=metrics.map(m=>`<div style="margin-bottom:14px"><div style="display:flex;justify-content:space-between;margin-bottom:5px"><span style="font-family:var(--mono);font-size:10px;color:var(--txt2)">${m.n}</span><span style="font-family:var(--display);font-size:12px;font-weight:700;color:${m.c}">${(m.v*100).toFixed(2)}%</span></div><div class="ptrack" style="margin:0;height:7px"><div class="pfill" style="width:${m.v*100}%;background:${m.c}"></div></div></div>`).join('');
654
+ const tot=d.total_transactions, tp=Math.round(d.fraud_detected*d.recall), fp2=Math.round(d.fraud_detected*(1-d.precision)), fn=d.fraud_detected-tp, tn=tot-tp-fp2-fn;
655
+ document.getElementById('cmTN').textContent=tn.toLocaleString();
656
+ document.getElementById('cmFP').textContent=fp2.toLocaleString();
657
+ document.getElementById('cmFN').textContent=fn.toLocaleString();
658
+ document.getElementById('cmTP').textContent=tp.toLocaleString();
659
+ const hrs=d.hourly_stats.map(h=>h.hour.slice(0,5));
660
+ lineChart('perfChart',hrs,[
661
+ {label:'Transactions',data:d.hourly_stats.map(h=>h.transactions),borderColor:'#00d4ff',backgroundColor:'rgba(0,212,255,.06)',tension:.4,fill:true,pointRadius:0},
662
+ {label:'Fraud',data:d.hourly_stats.map(h=>h.fraud),borderColor:'#ff2d55',backgroundColor:'rgba(255,45,85,.05)',tension:.4,fill:true,pointRadius:0}
663
+ ]);
664
+ updateThresh(0.5);
665
+ }
666
+
667
+ function updateThresh(v){
668
+ document.getElementById('threshVal').textContent=parseFloat(v).toFixed(2);
669
+ const prec=Math.min(.99,.7+v*.35).toFixed(3), rec=Math.max(.4,1-v*.7).toFixed(3);
670
+ const f1=(2*prec*rec/(+prec+ +rec)).toFixed(3);
671
+ document.getElementById('threshStats').innerHTML=`
672
+ <div class="mrow"><span class="mname">Precision</span><span class="mval">${prec}</span></div>
673
+ <div class="mrow"><span class="mname">Recall</span><span class="mval" style="color:var(--orange)">${rec}</span></div>
674
+ <div class="mrow"><span class="mname">F1 Score</span><span class="mval" style="color:var(--yellow)">${f1}</span></div>
675
+ <div class="mrow"><span class="mname">Recommendation</span><span style="font-family:var(--mono);font-size:9px;color:${+v>.6?'var(--orange)':+v<.3?'var(--red)':'var(--green)'}">${+v>.6?'High precision, may miss fraud':+v<.3?'High recall, many false alarms':'Balanced β€” recommended'}</span></div>`;
676
+ }
677
+
678
+ // ══════════════════════════════════════════════════════════
679
+ // SHAP
680
+ // ══════════════════════════════════════════════════════════
681
+ async function loadShap(){
682
+ const d=await get('/shap/global'); if(!d) return;
683
+ const feats=d.feature_importance.slice(0,15);
684
+ barChart('shapChart',feats.map(f=>f.feature),[{data:feats.map(f=>f.importance),backgroundColor:feats.map((_,i)=>i<5?'rgba(255,45,85,.7)':i<10?'rgba(255,149,0,.6)':'rgba(0,212,255,.5)'),borderWidth:0,borderRadius:3}],{indexAxis:'y'});
685
+ document.getElementById('topFeatures').innerHTML=feats.slice(0,7).map((f,i)=>`<div style="display:flex;align-items:center;gap:8px;padding:5px 0;border-bottom:1px solid rgba(21,40,64,.4)"><span style="font-family:var(--display);font-size:10px;color:var(--txt3);width:16px">#${i+1}</span><span style="font-family:var(--mono);font-size:11px;color:var(--cyan);width:36px">${f.feature}</span><div class="ptrack" style="margin:0;flex:1"><div class="pfill" style="width:${(f.importance/feats[0].importance*100).toFixed(0)}%;background:${i<3?'#ff2d55':'#00d4ff'}"></div></div><span style="font-family:var(--mono);font-size:9px;color:var(--txt3);width:46px;text-align:right">${f.importance.toFixed(4)}</span></div>`).join('');
686
+ }
687
+
688
+ function explainRandom(){
689
+ const fraud=Math.random()<.35;
690
+ const feats=['V14','V4','V12','V10','V17','V3','V11','V16','V7','Amount'];
691
+ const shapVals=fraud?[.35,.28,.18,.14,.12,-.05,-.04,.08,.07,.06]:[-.12,-.08,-.06,-.05,-.04,.08,.06,-.03,-.03,-.02];
692
+ const vals=fraud?[-7.2,5.1,-4.1,-3.3,-3.8,1.2,2.1,-1.9,-2.0,142]:[-.3,.1,.2,-.1,.0,.8,-.2,.3,-.1,1823];
693
+ const prob=fraud?(.65+Math.random()*.3):(.05+Math.random()*.2);
694
+ const color=fraud?'var(--red)':'var(--green)';
695
+ document.getElementById('localShap').innerHTML=`
696
+ <div style="display:flex;align-items:center;gap:14px;margin-bottom:16px;padding:14px;background:var(--bg2);border-radius:7px;border:1px solid ${fraud?'rgba(255,45,85,.3)':'rgba(0,255,136,.2)'}">
697
+ <div style="font-size:32px">${fraud?'⚠':'βœ“'}</div>
698
+ <div>
699
+ <div style="font-family:var(--display);font-size:14px;font-weight:900;color:${color}">${fraud?'FRAUD DETECTED':'TRANSACTION CLEAR'}</div>
700
+ <div style="font-family:var(--mono);font-size:9px;color:var(--txt3)">Fraud Probability: ${(prob*100).toFixed(1)}% β€’ Amount: β‚Ή${vals[9].toFixed(2)}</div>
701
+ </div>
702
+ <div style="margin-left:auto;text-align:right">
703
+ <div style="font-family:var(--display);font-size:24px;font-weight:900;color:${color}">${(prob*100).toFixed(1)}%</div>
704
+ <div style="font-family:var(--mono);font-size:8px;color:var(--txt3)">FRAUD SCORE</div>
705
+ </div>
706
+ </div>
707
+ <div style="font-family:var(--mono);font-size:8px;color:var(--txt3);letter-spacing:2px;margin-bottom:10px">FEATURE CONTRIBUTIONS</div>
708
+ ${feats.map((f,i)=>{ const sv=shapVals[i]*(0.8+Math.random()*.4); const pct=Math.min(Math.abs(sv)/.4*50,50); return `<div class="shap-row"><span class="shap-f">${f}</span><div class="shap-bc"><div class="${sv>0?'shap-pos':'shap-neg'}" style="width:${pct}%"></div></div><span class="shap-v" style="color:${sv>0?'var(--red)':'var(--cyan)'}">${sv>0?'+':''}${sv.toFixed(4)}</span></div>`; }).join('')}
709
+ <div style="margin-top:14px;font-family:var(--mono);font-size:9px;color:var(--txt3);padding:10px;background:var(--bg2);border-radius:5px;line-height:1.8">
710
+ ${fraud?`<span style="color:var(--red)">⚠ FRAUD:</span> Primarily flagged due to V14 (${vals[0].toFixed(2)}) β€” highly negative, strong fraud signal. Combined with elevated V4 and V12, model is confident.`:`<span style="color:var(--green)">βœ“ LEGIT:</span> V14 and key features are within normal ranges. High amount (β‚Ή${vals[9].toFixed(0)}) is not suspicious given normal feature pattern.`}
711
+ </div>`;
712
+ }
713
+
714
+ // ── Init ───────────────────────────────────────────────────
715
+ loadOverview();
716
+ </script>
717
+ </body>
718
+ </html>
model/amount_scaler.pkl ADDED
Binary file (927 Bytes). View file
 
model/feature_names.pkl ADDED
Binary file (184 Bytes). View file
 
model/test_sample.csv ADDED
The diff for this file is too large to render. See raw diff
 
notebooks/01_eda_and_training.ipynb ADDED
@@ -0,0 +1,436 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cells": [
3
+ {
4
+ "cell_type": "markdown",
5
+ "metadata": {},
6
+ "source": [
7
+ "# πŸ” Real-Time Fraud Detection β€” EDA & Model Training\n",
8
+ "**Dataset**: [Kaggle Credit Card Fraud](https://www.kaggle.com/datasets/mlg-ulb/creditcardfraud) \n",
9
+ "**Model**: XGBoost with SMOTE + SHAP Explainability"
10
+ ]
11
+ },
12
+ {
13
+ "cell_type": "markdown",
14
+ "metadata": {},
15
+ "source": ["## 1️⃣ Imports & Setup"]
16
+ },
17
+ {
18
+ "cell_type": "code",
19
+ "execution_count": null,
20
+ "metadata": {},
21
+ "outputs": [],
22
+ "source": [
23
+ "import pandas as pd\n",
24
+ "import numpy as np\n",
25
+ "import matplotlib.pyplot as plt\n",
26
+ "import seaborn as sns\n",
27
+ "import warnings\n",
28
+ "warnings.filterwarnings('ignore')\n",
29
+ "\n",
30
+ "from sklearn.model_selection import train_test_split, StratifiedKFold, cross_val_score\n",
31
+ "from sklearn.preprocessing import StandardScaler\n",
32
+ "from sklearn.metrics import (classification_report, confusion_matrix,\n",
33
+ " roc_auc_score, precision_recall_curve,\n",
34
+ " average_precision_score, f1_score)\n",
35
+ "from imblearn.over_sampling import SMOTE\n",
36
+ "from xgboost import XGBClassifier\n",
37
+ "import shap\n",
38
+ "import joblib\n",
39
+ "import os\n",
40
+ "\n",
41
+ "sns.set_theme(style='whitegrid')\n",
42
+ "plt.rcParams['figure.figsize'] = (10, 5)\n",
43
+ "print('βœ… Libraries loaded!')"
44
+ ]
45
+ },
46
+ {
47
+ "cell_type": "markdown",
48
+ "metadata": {},
49
+ "source": ["## 2️⃣ Load Dataset\n", "> Download from: https://www.kaggle.com/datasets/mlg-ulb/creditcardfraud \n", "> Place `creditcard.csv` in the `data/` folder"]
50
+ },
51
+ {
52
+ "cell_type": "code",
53
+ "execution_count": null,
54
+ "metadata": {},
55
+ "outputs": [],
56
+ "source": [
57
+ "DATA_PATH = '../data/creditcard.csv'\n",
58
+ "\n",
59
+ "# ── If you don't have the file yet, generate synthetic sample data ──\n",
60
+ "if not os.path.exists(DATA_PATH):\n",
61
+ " print('⚠️ creditcard.csv not found β€” generating synthetic demo data...')\n",
62
+ " np.random.seed(42)\n",
63
+ " n = 10000\n",
64
+ " # 28 PCA features V1–V28 + Time + Amount + Class\n",
65
+ " V_cols = {f'V{i}': np.random.randn(n) for i in range(1, 29)}\n",
66
+ " df = pd.DataFrame(V_cols)\n",
67
+ " df['Time'] = np.random.uniform(0, 172800, n)\n",
68
+ " df['Amount'] = np.abs(np.random.exponential(88, n))\n",
69
+ " # Inject 2% fraud β€” make frauds slightly separable\n",
70
+ " fraud_idx = np.random.choice(n, size=200, replace=False)\n",
71
+ " df['Class'] = 0\n",
72
+ " df.loc[fraud_idx, 'Class'] = 1\n",
73
+ " df.loc[fraud_idx, 'V1'] -= 3\n",
74
+ " df.loc[fraud_idx, 'V3'] += 2\n",
75
+ " df.to_csv(DATA_PATH, index=False)\n",
76
+ " print(f'βœ… Synthetic data saved β†’ {DATA_PATH}')\n",
77
+ "\n",
78
+ "df = pd.read_csv(DATA_PATH)\n",
79
+ "print(f'Shape: {df.shape}')\n",
80
+ "df.head()"
81
+ ]
82
+ },
83
+ {
84
+ "cell_type": "markdown",
85
+ "metadata": {},
86
+ "source": ["## 3️⃣ Exploratory Data Analysis"]
87
+ },
88
+ {
89
+ "cell_type": "code",
90
+ "execution_count": null,
91
+ "metadata": {},
92
+ "outputs": [],
93
+ "source": [
94
+ "# ── Basic Info ──\n",
95
+ "print('=== Dataset Info ===')\n",
96
+ "print(f'Total transactions : {len(df):,}')\n",
97
+ "print(f'Fraudulent : {df[\"Class\"].sum():,} ({df[\"Class\"].mean()*100:.3f}%)')\n",
98
+ "print(f'Legitimate : {(df[\"Class\"]==0).sum():,}')\n",
99
+ "print(f'Missing values : {df.isnull().sum().sum()}')\n",
100
+ "print()\n",
101
+ "df.describe()"
102
+ ]
103
+ },
104
+ {
105
+ "cell_type": "code",
106
+ "execution_count": null,
107
+ "metadata": {},
108
+ "outputs": [],
109
+ "source": [
110
+ "# ── Class Imbalance ──\n",
111
+ "fig, axes = plt.subplots(1, 2, figsize=(12, 4))\n",
112
+ "\n",
113
+ "# Pie chart\n",
114
+ "counts = df['Class'].value_counts()\n",
115
+ "axes[0].pie(counts, labels=['Legitimate', 'Fraud'], autopct='%1.2f%%',\n",
116
+ " colors=['#4CAF50', '#F44336'], startangle=90,\n",
117
+ " wedgeprops=dict(edgecolor='white', linewidth=2))\n",
118
+ "axes[0].set_title('Class Distribution', fontsize=14, fontweight='bold')\n",
119
+ "\n",
120
+ "# Amount distribution by class\n",
121
+ "df[df['Class']==0]['Amount'].hist(bins=60, ax=axes[1], alpha=0.6, color='#4CAF50', label='Legitimate')\n",
122
+ "df[df['Class']==1]['Amount'].hist(bins=60, ax=axes[1], alpha=0.7, color='#F44336', label='Fraud')\n",
123
+ "axes[1].set_xlabel('Transaction Amount (€)')\n",
124
+ "axes[1].set_ylabel('Frequency')\n",
125
+ "axes[1].set_title('Amount Distribution by Class', fontsize=14, fontweight='bold')\n",
126
+ "axes[1].legend()\n",
127
+ "axes[1].set_xlim(0, 1000)\n",
128
+ "\n",
129
+ "plt.tight_layout()\n",
130
+ "plt.savefig('../data/eda_class_distribution.png', dpi=120, bbox_inches='tight')\n",
131
+ "plt.show()\n",
132
+ "print('πŸ’‘ Key Insight: Massive class imbalance β†’ need SMOTE!')"
133
+ ]
134
+ },
135
+ {
136
+ "cell_type": "code",
137
+ "execution_count": null,
138
+ "metadata": {},
139
+ "outputs": [],
140
+ "source": [
141
+ "# ── Feature Correlation with Fraud ──\n",
142
+ "correlations = df.corr()['Class'].drop('Class').sort_values()\n",
143
+ "\n",
144
+ "fig, ax = plt.subplots(figsize=(12, 6))\n",
145
+ "colors = ['#F44336' if x > 0 else '#2196F3' for x in correlations]\n",
146
+ "correlations.plot(kind='bar', ax=ax, color=colors, edgecolor='white', linewidth=0.5)\n",
147
+ "ax.axhline(0, color='black', linewidth=0.8)\n",
148
+ "ax.set_title('Feature Correlation with Fraud Class', fontsize=14, fontweight='bold')\n",
149
+ "ax.set_xlabel('Feature')\n",
150
+ "ax.set_ylabel('Pearson Correlation')\n",
151
+ "plt.xticks(rotation=45, ha='right')\n",
152
+ "plt.tight_layout()\n",
153
+ "plt.savefig('../data/eda_correlations.png', dpi=120, bbox_inches='tight')\n",
154
+ "plt.show()\n",
155
+ "print('Top positive correlations (fraud indicators):')\n",
156
+ "print(correlations.tail(5))"
157
+ ]
158
+ },
159
+ {
160
+ "cell_type": "code",
161
+ "execution_count": null,
162
+ "metadata": {},
163
+ "outputs": [],
164
+ "source": [
165
+ "# ── Top features violin plots ──\n",
166
+ "top_features = correlations.abs().sort_values(ascending=False).index[:6].tolist()\n",
167
+ "\n",
168
+ "fig, axes = plt.subplots(2, 3, figsize=(14, 8))\n",
169
+ "axes = axes.flatten()\n",
170
+ "\n",
171
+ "for i, feat in enumerate(top_features):\n",
172
+ " data_plot = [\n",
173
+ " df[df['Class']==0][feat].values,\n",
174
+ " df[df['Class']==1][feat].values\n",
175
+ " ]\n",
176
+ " parts = axes[i].violinplot(data_plot, showmedians=True)\n",
177
+ " for j, pc in enumerate(parts['bodies']):\n",
178
+ " pc.set_facecolor('#4CAF50' if j==0 else '#F44336')\n",
179
+ " pc.set_alpha(0.7)\n",
180
+ " axes[i].set_title(feat, fontsize=12, fontweight='bold')\n",
181
+ " axes[i].set_xticks([1, 2])\n",
182
+ " axes[i].set_xticklabels(['Legit', 'Fraud'])\n",
183
+ " axes[i].set_ylabel('Value')\n",
184
+ "\n",
185
+ "plt.suptitle('Top Feature Distributions: Legit vs Fraud', fontsize=14, fontweight='bold', y=1.01)\n",
186
+ "plt.tight_layout()\n",
187
+ "plt.savefig('../data/eda_violin.png', dpi=120, bbox_inches='tight')\n",
188
+ "plt.show()"
189
+ ]
190
+ },
191
+ {
192
+ "cell_type": "markdown",
193
+ "metadata": {},
194
+ "source": ["## 4️⃣ Feature Engineering & Preprocessing"]
195
+ },
196
+ {
197
+ "cell_type": "code",
198
+ "execution_count": null,
199
+ "metadata": {},
200
+ "outputs": [],
201
+ "source": [
202
+ "# ── Scale Amount (Time is dropped β€” not predictive) ──\n",
203
+ "df_processed = df.drop(columns=['Time']).copy()\n",
204
+ "\n",
205
+ "scaler = StandardScaler()\n",
206
+ "df_processed['Amount'] = scaler.fit_transform(df_processed[['Amount']])\n",
207
+ "\n",
208
+ "X = df_processed.drop('Class', axis=1)\n",
209
+ "y = df_processed['Class']\n",
210
+ "\n",
211
+ "print(f'Features: {X.shape[1]}')\n",
212
+ "print(f'Class balance before SMOTE: {y.value_counts().to_dict()}')"
213
+ ]
214
+ },
215
+ {
216
+ "cell_type": "code",
217
+ "execution_count": null,
218
+ "metadata": {},
219
+ "outputs": [],
220
+ "source": [
221
+ "# ── Train/Test Split ──\n",
222
+ "X_train, X_test, y_train, y_test = train_test_split(\n",
223
+ " X, y, test_size=0.2, random_state=42, stratify=y\n",
224
+ ")\n",
225
+ "\n",
226
+ "# ── Apply SMOTE on training set only ──\n",
227
+ "smote = SMOTE(random_state=42)\n",
228
+ "X_train_res, y_train_res = smote.fit_resample(X_train, y_train)\n",
229
+ "\n",
230
+ "print(f'Before SMOTE β†’ Train: {dict(pd.Series(y_train).value_counts())}')\n",
231
+ "print(f'After SMOTE β†’ Train: {dict(pd.Series(y_train_res).value_counts())}')\n",
232
+ "print(f'Test set (no SMOTE): {dict(pd.Series(y_test).value_counts())}')"
233
+ ]
234
+ },
235
+ {
236
+ "cell_type": "markdown",
237
+ "metadata": {},
238
+ "source": ["## 5️⃣ Model Training β€” XGBoost"]
239
+ },
240
+ {
241
+ "cell_type": "code",
242
+ "execution_count": null,
243
+ "metadata": {},
244
+ "outputs": [],
245
+ "source": [
246
+ "# ── Train XGBoost ──\n",
247
+ "model = XGBClassifier(\n",
248
+ " n_estimators=300,\n",
249
+ " max_depth=6,\n",
250
+ " learning_rate=0.05,\n",
251
+ " subsample=0.8,\n",
252
+ " colsample_bytree=0.8,\n",
253
+ " min_child_weight=1,\n",
254
+ " gamma=0.1,\n",
255
+ " reg_alpha=0.1,\n",
256
+ " reg_lambda=1,\n",
257
+ " use_label_encoder=False,\n",
258
+ " eval_metric='logloss',\n",
259
+ " random_state=42,\n",
260
+ " n_jobs=-1\n",
261
+ ")\n",
262
+ "\n",
263
+ "model.fit(\n",
264
+ " X_train_res, y_train_res,\n",
265
+ " eval_set=[(X_test, y_test)],\n",
266
+ " verbose=50\n",
267
+ ")\n",
268
+ "print('\\nβœ… Model trained!')"
269
+ ]
270
+ },
271
+ {
272
+ "cell_type": "markdown",
273
+ "metadata": {},
274
+ "source": ["## 6️⃣ Model Evaluation"]
275
+ },
276
+ {
277
+ "cell_type": "code",
278
+ "execution_count": null,
279
+ "metadata": {},
280
+ "outputs": [],
281
+ "source": [
282
+ "# ── Predictions ──\n",
283
+ "y_pred = model.predict(X_test)\n",
284
+ "y_pred_proba = model.predict_proba(X_test)[:, 1]\n",
285
+ "\n",
286
+ "# ── Metrics ──\n",
287
+ "print('=== Classification Report ===')\n",
288
+ "print(classification_report(y_test, y_pred, target_names=['Legit', 'Fraud']))\n",
289
+ "\n",
290
+ "roc_auc = roc_auc_score(y_test, y_pred_proba)\n",
291
+ "pr_auc = average_precision_score(y_test, y_pred_proba)\n",
292
+ "f1 = f1_score(y_test, y_pred)\n",
293
+ "print(f'ROC-AUC Score : {roc_auc:.4f}')\n",
294
+ "print(f'PR-AUC Score : {pr_auc:.4f}')\n",
295
+ "print(f'F1 Score (Fraud) : {f1:.4f}')"
296
+ ]
297
+ },
298
+ {
299
+ "cell_type": "code",
300
+ "execution_count": null,
301
+ "metadata": {},
302
+ "outputs": [],
303
+ "source": [
304
+ "# ── Confusion Matrix ──\n",
305
+ "fig, axes = plt.subplots(1, 2, figsize=(12, 5))\n",
306
+ "\n",
307
+ "cm = confusion_matrix(y_test, y_pred)\n",
308
+ "sns.heatmap(cm, annot=True, fmt='d', cmap='Blues',\n",
309
+ " xticklabels=['Legit', 'Fraud'],\n",
310
+ " yticklabels=['Legit', 'Fraud'], ax=axes[0],\n",
311
+ " linewidths=1, linecolor='white')\n",
312
+ "axes[0].set_title('Confusion Matrix', fontsize=13, fontweight='bold')\n",
313
+ "axes[0].set_ylabel('Actual')\n",
314
+ "axes[0].set_xlabel('Predicted')\n",
315
+ "\n",
316
+ "# Precision-Recall Curve\n",
317
+ "precision, recall, thresholds = precision_recall_curve(y_test, y_pred_proba)\n",
318
+ "axes[1].plot(recall, precision, color='#E91E63', linewidth=2.5, label=f'PR-AUC = {pr_auc:.4f}')\n",
319
+ "axes[1].fill_between(recall, precision, alpha=0.1, color='#E91E63')\n",
320
+ "axes[1].set_xlabel('Recall')\n",
321
+ "axes[1].set_ylabel('Precision')\n",
322
+ "axes[1].set_title('Precision-Recall Curve', fontsize=13, fontweight='bold')\n",
323
+ "axes[1].legend(fontsize=12)\n",
324
+ "axes[1].grid(True, alpha=0.3)\n",
325
+ "\n",
326
+ "plt.tight_layout()\n",
327
+ "plt.savefig('../data/evaluation_plots.png', dpi=120, bbox_inches='tight')\n",
328
+ "plt.show()"
329
+ ]
330
+ },
331
+ {
332
+ "cell_type": "markdown",
333
+ "metadata": {},
334
+ "source": ["## 7️⃣ SHAP Explainability"]
335
+ },
336
+ {
337
+ "cell_type": "code",
338
+ "execution_count": null,
339
+ "metadata": {},
340
+ "outputs": [],
341
+ "source": [
342
+ "# ── Compute SHAP Values ──\n",
343
+ "print('Computing SHAP values...')\n",
344
+ "explainer = shap.TreeExplainer(model)\n",
345
+ "shap_values = explainer.shap_values(X_test)\n",
346
+ "\n",
347
+ "# ── Global Feature Importance ──\n",
348
+ "plt.figure(figsize=(10, 7))\n",
349
+ "shap.summary_plot(shap_values, X_test, plot_type='bar', show=False)\n",
350
+ "plt.title('SHAP Global Feature Importance', fontsize=14, fontweight='bold')\n",
351
+ "plt.tight_layout()\n",
352
+ "plt.savefig('../data/shap_importance.png', dpi=120, bbox_inches='tight')\n",
353
+ "plt.show()"
354
+ ]
355
+ },
356
+ {
357
+ "cell_type": "code",
358
+ "execution_count": null,
359
+ "metadata": {},
360
+ "outputs": [],
361
+ "source": [
362
+ "# ── SHAP Beeswarm (how each feature affects fraud score) ──\n",
363
+ "plt.figure(figsize=(10, 8))\n",
364
+ "shap.summary_plot(shap_values, X_test, show=False)\n",
365
+ "plt.title('SHAP Summary β€” Feature Impact on Fraud Probability', fontsize=13, fontweight='bold')\n",
366
+ "plt.tight_layout()\n",
367
+ "plt.savefig('../data/shap_beeswarm.png', dpi=120, bbox_inches='tight')\n",
368
+ "plt.show()"
369
+ ]
370
+ },
371
+ {
372
+ "cell_type": "code",
373
+ "execution_count": null,
374
+ "metadata": {},
375
+ "outputs": [],
376
+ "source": [
377
+ "# ── SHAP Waterfall for ONE fraud case (explainability for single transaction) ──\n",
378
+ "fraud_indices = np.where(y_test == 1)[0]\n",
379
+ "if len(fraud_indices) > 0:\n",
380
+ " sample_idx = fraud_indices[0]\n",
381
+ " print(f'Explaining fraud transaction at test index: {sample_idx}')\n",
382
+ " explanation = explainer(X_test.iloc[[sample_idx]])\n",
383
+ " plt.figure(figsize=(10, 6))\n",
384
+ " shap.waterfall_plot(explanation[0], show=False)\n",
385
+ " plt.title('Why was this transaction flagged as FRAUD?', fontsize=13, fontweight='bold')\n",
386
+ " plt.tight_layout()\n",
387
+ " plt.savefig('../data/shap_waterfall_fraud.png', dpi=120, bbox_inches='tight')\n",
388
+ " plt.show()"
389
+ ]
390
+ },
391
+ {
392
+ "cell_type": "markdown",
393
+ "metadata": {},
394
+ "source": ["## 8️⃣ Save Model Artifacts"]
395
+ },
396
+ {
397
+ "cell_type": "code",
398
+ "execution_count": null,
399
+ "metadata": {},
400
+ "outputs": [],
401
+ "source": [
402
+ "os.makedirs('../model', exist_ok=True)\n",
403
+ "\n",
404
+ "# Save model\n",
405
+ "joblib.dump(model, '../model/xgb_fraud_model.pkl')\n",
406
+ "# Save scaler\n",
407
+ "joblib.dump(scaler, '../model/amount_scaler.pkl')\n",
408
+ "# Save feature names\n",
409
+ "joblib.dump(list(X.columns), '../model/feature_names.pkl')\n",
410
+ "# Save test sample for demo\n",
411
+ "X_test.to_csv('../model/test_sample.csv', index=False)\n",
412
+ "\n",
413
+ "print('βœ… Model artifacts saved:')\n",
414
+ "print(' β†’ model/xgb_fraud_model.pkl')\n",
415
+ "print(' β†’ model/amount_scaler.pkl')\n",
416
+ "print(' β†’ model/feature_names.pkl')\n",
417
+ "print(' β†’ model/test_sample.csv')\n",
418
+ "print(f'\\nFinal ROC-AUC: {roc_auc:.4f}')\n",
419
+ "print(f'Final PR-AUC : {pr_auc:.4f}')"
420
+ ]
421
+ }
422
+ ],
423
+ "metadata": {
424
+ "kernelspec": {
425
+ "display_name": "Python 3",
426
+ "language": "python",
427
+ "name": "python3"
428
+ },
429
+ "language_info": {
430
+ "name": "python",
431
+ "version": "3.10.0"
432
+ }
433
+ },
434
+ "nbformat": 4,
435
+ "nbformat_minor": 5
436
+ }
notebooks/creditcard.csv ADDED
The diff for this file is too large to render. See raw diff
 
requirements.txt ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Core ML
2
+ pandas==2.1.4
3
+ numpy==1.26.2
4
+ scikit-learn==1.3.2
5
+ xgboost==2.0.3
6
+ imbalanced-learn==0.11.0
7
+ joblib==1.3.2
8
+
9
+ # Explainability
10
+ shap==0.44.1
11
+
12
+ # API
13
+ fastapi==0.108.0
14
+ uvicorn==0.25.0
15
+ pydantic==2.5.3
16
+
17
+ # Kafka simulation
18
+ kafka-python==2.0.2
19
+
20
+ # Dashboard
21
+ streamlit==1.29.0
22
+ plotly==5.18.0
23
+ requests==2.31.0
24
+
25
+ # Notebook
26
+ jupyter==1.0.0
27
+ matplotlib==3.8.2
28
+ seaborn==0.13.0
stream/consumer.py ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Kafka Consumer β€” Reads live transactions and scores them via FastAPI
3
+ Logs fraud alerts in real time
4
+ """
5
+
6
+ import json
7
+ import time
8
+ import requests
9
+ from kafka import KafkaConsumer
10
+ import logging
11
+
12
+ logging.basicConfig(level=logging.INFO, format='%(asctime)s | %(message)s')
13
+ logger = logging.getLogger(__name__)
14
+
15
+ KAFKA_BROKER = "localhost:9092"
16
+ TOPIC = "transactions"
17
+ API_URL = "http://localhost:8000/predict"
18
+ GROUP_ID = "fraud-detection-group"
19
+
20
+
21
+ def score_transaction(txn: dict) -> dict | None:
22
+ """Call FastAPI scoring endpoint for a single transaction"""
23
+ payload = {k: v for k, v in txn.items()
24
+ if k.startswith("V") or k == "Amount"}
25
+ try:
26
+ resp = requests.post(API_URL, json=payload, timeout=2)
27
+ return resp.json() if resp.status_code == 200 else None
28
+ except Exception as e:
29
+ logger.error(f"API call failed: {e}")
30
+ return None
31
+
32
+
33
+ def main():
34
+ logger.info(f"Connecting to Kafka broker at {KAFKA_BROKER}...")
35
+ try:
36
+ consumer = KafkaConsumer(
37
+ TOPIC,
38
+ bootstrap_servers=[KAFKA_BROKER],
39
+ group_id=GROUP_ID,
40
+ value_deserializer=lambda m: json.loads(m.decode("utf-8")),
41
+ auto_offset_reset="latest",
42
+ enable_auto_commit=True
43
+ )
44
+ logger.info(f"βœ… Listening on topic: [{TOPIC}]")
45
+ except Exception as e:
46
+ logger.error(f"❌ Cannot connect to Kafka: {e}")
47
+ return
48
+
49
+ total = fraud = 0
50
+ try:
51
+ for msg in consumer:
52
+ txn = msg.value
53
+ result = score_transaction(txn)
54
+
55
+ total += 1
56
+ if result and result.get("is_fraud"):
57
+ fraud += 1
58
+ logger.warning(
59
+ f"🚨 FRAUD ALERT! | TXN: {result['transaction_id']} "
60
+ f"| Prob: {result['fraud_probability']:.1%} "
61
+ f"| Risk: {result['risk_level']} "
62
+ f"| Amount: ${txn.get('Amount', 0):.2f}"
63
+ )
64
+ logger.warning(
65
+ f" Top factor: {result['top_risk_factors'][0]['feature']} "
66
+ f"(SHAP={result['top_risk_factors'][0]['shap_value']})"
67
+ )
68
+ else:
69
+ if total % 10 == 0: # log every 10th legit
70
+ logger.info(f"βœ… LEGIT | Total processed: {total} | Frauds caught: {fraud}")
71
+
72
+ except KeyboardInterrupt:
73
+ logger.info(f"\nβ›” Stopped. Total={total} | Frauds={fraud} | Rate={fraud/max(total,1)*100:.1f}%")
74
+ consumer.close()
75
+
76
+
77
+ if __name__ == "__main__":
78
+ main()
stream/producer.py ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Kafka Producer β€” Simulates a real-time transaction stream
3
+ Sends random transactions (some seeded as fraud) to the Kafka topic 'transactions'
4
+ """
5
+
6
+ import json
7
+ import time
8
+ import random
9
+ import numpy as np
10
+ from kafka import KafkaProducer
11
+ import logging
12
+
13
+ logging.basicConfig(level=logging.INFO, format='%(asctime)s | %(message)s')
14
+ logger = logging.getLogger(__name__)
15
+
16
+ KAFKA_BROKER = "localhost:9092"
17
+ TOPIC = "transactions"
18
+ SEND_RATE_SEC = 0.5 # one transaction every 0.5s
19
+ FRAUD_RATE = 0.05 # 5% of messages are fraud-like
20
+
21
+
22
+ def generate_transaction(force_fraud: bool = False) -> dict:
23
+ """Generate a synthetic transaction matching creditcard.csv schema"""
24
+ np.random.seed(None)
25
+
26
+ if force_fraud:
27
+ # Fraud fingerprint β€” V1 very negative, high V3
28
+ txn = {f"V{i}": np.random.randn() for i in range(1, 29)}
29
+ txn["V1"] -= random.uniform(2, 5)
30
+ txn["V3"] += random.uniform(2, 4)
31
+ txn["V14"] -= random.uniform(2, 4)
32
+ txn["Amount"] = random.uniform(1, 50) # small amounts typical in fraud
33
+ else:
34
+ txn = {f"V{i}": np.random.randn() for i in range(1, 29)}
35
+ txn["Amount"] = abs(np.random.exponential(88))
36
+
37
+ # Round to 4 dp
38
+ txn = {k: round(float(v), 4) for k, v in txn.items()}
39
+ txn["timestamp"] = time.time()
40
+ txn["is_simulated_fraud"] = force_fraud # ground-truth label for evaluation
41
+ return txn
42
+
43
+
44
+ def on_send_success(metadata):
45
+ logger.info(f"βœ… Sent to [{metadata.topic}] partition={metadata.partition} offset={metadata.offset}")
46
+
47
+
48
+ def on_send_error(exc):
49
+ logger.error(f"❌ Send failed: {exc}")
50
+
51
+
52
+ def main():
53
+ logger.info(f"Connecting to Kafka broker at {KAFKA_BROKER}...")
54
+ try:
55
+ producer = KafkaProducer(
56
+ bootstrap_servers=[KAFKA_BROKER],
57
+ value_serializer=lambda v: json.dumps(v).encode("utf-8"),
58
+ acks="all",
59
+ retries=3
60
+ )
61
+ logger.info(f"βœ… Connected! Streaming to topic: [{TOPIC}]")
62
+ except Exception as e:
63
+ logger.error(f"❌ Cannot connect to Kafka: {e}")
64
+ logger.info("πŸ’‘ Start Kafka first:\n docker-compose up kafka zookeeper")
65
+ return
66
+
67
+ count = 0
68
+ try:
69
+ while True:
70
+ is_fraud = random.random() < FRAUD_RATE
71
+ txn = generate_transaction(force_fraud=is_fraud)
72
+
73
+ producer.send(TOPIC, value=txn).add_callback(on_send_success).add_errback(on_send_error)
74
+ producer.flush()
75
+
76
+ count += 1
77
+ tag = "🚨 FRAUD" if is_fraud else "βœ… LEGIT"
78
+ logger.info(f"[#{count}] {tag} | Amount=${txn['Amount']:.2f}")
79
+
80
+ time.sleep(SEND_RATE_SEC)
81
+
82
+ except KeyboardInterrupt:
83
+ logger.info(f"\nβ›” Stopped. Total sent: {count}")
84
+ producer.close()
85
+
86
+
87
+ if __name__ == "__main__":
88
+ main()
utils.js ADDED
@@ -0,0 +1,221 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ const API_BASE = 'http://localhost:8000';
2
+
3
+ // ── API helpers ────────────────────────────────────────────
4
+ async function apiFetch(path, options = {}) {
5
+ try {
6
+ const res = await fetch(API_BASE + path, options);
7
+ if (!res.ok) throw new Error(`HTTP ${res.status}`);
8
+ return await res.json();
9
+ } catch (e) {
10
+ console.error('API error:', e);
11
+ return null;
12
+ }
13
+ }
14
+
15
+ // ── Clock ─────────────────────────────────────────────────
16
+ function startClock(el) {
17
+ function tick() {
18
+ const now = new Date();
19
+ el.textContent = now.toLocaleTimeString('en-IN', { hour12: false }) +
20
+ ' IST';
21
+ }
22
+ tick();
23
+ setInterval(tick, 1000);
24
+ }
25
+
26
+ // ── Format helpers ─────────────────────────────────────────
27
+ function fmtCurrency(n) {
28
+ if (n >= 1e6) return 'β‚Ή' + (n / 1e6).toFixed(2) + 'M';
29
+ if (n >= 1e3) return 'β‚Ή' + (n / 1e3).toFixed(1) + 'K';
30
+ return 'β‚Ή' + n.toFixed(2);
31
+ }
32
+
33
+ function fmtPct(n) { return (n * 100).toFixed(2) + '%'; }
34
+
35
+ function riskBadge(level) {
36
+ const map = { HIGH: 'badge-high', MEDIUM: 'badge-medium', LOW: 'badge-low' };
37
+ return `<span class="badge ${map[level] || 'badge-low'}">${level}</span>`;
38
+ }
39
+
40
+ function fraudBadge(isFraud) {
41
+ return isFraud
42
+ ? `<span class="badge badge-fraud">⚠ FRAUD</span>`
43
+ : `<span class="badge badge-legit">βœ“ LEGIT</span>`;
44
+ }
45
+
46
+ function probBar(prob) {
47
+ const pct = (prob * 100).toFixed(0);
48
+ const color = prob > 0.7 ? '#ff2d55' : prob > 0.4 ? '#ff9500' : '#00ff88';
49
+ return `<span class="prob-bar"><span class="prob-fill" style="width:${pct}%;background:${color}"></span></span>${pct}%`;
50
+ }
51
+
52
+ function timeAgo(ts) {
53
+ const diff = Math.floor((Date.now() - new Date(ts)) / 1000);
54
+ if (diff < 60) return diff + 's ago';
55
+ if (diff < 3600) return Math.floor(diff / 60) + 'm ago';
56
+ return Math.floor(diff / 3600) + 'h ago';
57
+ }
58
+
59
+ // ── Sidebar active link ───────────────────────────────────
60
+ function setActiveNav() {
61
+ const path = window.location.pathname;
62
+ document.querySelectorAll('.nav-item').forEach(a => {
63
+ a.classList.remove('active');
64
+ if (a.getAttribute('href') === path) a.classList.add('active');
65
+ });
66
+ }
67
+
68
+ // ── Mini sparkline canvas ─────────────────────────────────
69
+ function drawSparkline(canvas, data, color = '#00d4ff') {
70
+ const ctx = canvas.getContext('2d');
71
+ const w = canvas.width, h = canvas.height;
72
+ const max = Math.max(...data), min = Math.min(...data);
73
+ const range = max - min || 1;
74
+ ctx.clearRect(0, 0, w, h);
75
+ ctx.beginPath();
76
+ data.forEach((v, i) => {
77
+ const x = (i / (data.length - 1)) * w;
78
+ const y = h - ((v - min) / range) * (h - 4) - 2;
79
+ i === 0 ? ctx.moveTo(x, y) : ctx.lineTo(x, y);
80
+ });
81
+ ctx.strokeStyle = color;
82
+ ctx.lineWidth = 1.5;
83
+ ctx.stroke();
84
+
85
+ // Fill gradient
86
+ ctx.lineTo(w, h); ctx.lineTo(0, h); ctx.closePath();
87
+ const grad = ctx.createLinearGradient(0, 0, 0, h);
88
+ grad.addColorStop(0, color + '33');
89
+ grad.addColorStop(1, color + '00');
90
+ ctx.fillStyle = grad;
91
+ ctx.fill();
92
+ }
93
+
94
+ // ── Generic Chart.js bar ──────────────────────────────────
95
+ function drawBarChart(canvasId, labels, datasets, options = {}) {
96
+ const ctx = document.getElementById(canvasId);
97
+ if (!ctx) return;
98
+ if (ctx._chart) ctx._chart.destroy();
99
+ ctx._chart = new Chart(ctx, {
100
+ type: 'bar',
101
+ data: { labels, datasets },
102
+ options: {
103
+ responsive: true,
104
+ maintainAspectRatio: false,
105
+ plugins: { legend: { display: false }, ...options.plugins },
106
+ scales: {
107
+ x: {
108
+ grid: { color: '#1a3a5c55' },
109
+ ticks: { color: '#7fa8c9', font: { family: 'Share Tech Mono', size: 9 } }
110
+ },
111
+ y: {
112
+ grid: { color: '#1a3a5c55' },
113
+ ticks: { color: '#7fa8c9', font: { family: 'Share Tech Mono', size: 9 } }
114
+ }
115
+ },
116
+ ...options
117
+ }
118
+ });
119
+ return ctx._chart;
120
+ }
121
+
122
+ function drawLineChart(canvasId, labels, datasets, options = {}) {
123
+ const ctx = document.getElementById(canvasId);
124
+ if (!ctx) return;
125
+ if (ctx._chart) ctx._chart.destroy();
126
+ ctx._chart = new Chart(ctx, {
127
+ type: 'line',
128
+ data: { labels, datasets },
129
+ options: {
130
+ responsive: true,
131
+ maintainAspectRatio: false,
132
+ plugins: { legend: { labels: { color: '#7fa8c9', font: { family: 'Share Tech Mono', size: 10 } } }, ...options.plugins },
133
+ scales: {
134
+ x: {
135
+ grid: { color: '#1a3a5c44' },
136
+ ticks: { color: '#7fa8c9', font: { family: 'Share Tech Mono', size: 9 } }
137
+ },
138
+ y: {
139
+ grid: { color: '#1a3a5c44' },
140
+ ticks: { color: '#7fa8c9', font: { family: 'Share Tech Mono', size: 9 } }
141
+ }
142
+ },
143
+ ...options
144
+ }
145
+ });
146
+ return ctx._chart;
147
+ }
148
+
149
+ function drawDoughnut(canvasId, labels, data, colors) {
150
+ const ctx = document.getElementById(canvasId);
151
+ if (!ctx) return;
152
+ if (ctx._chart) ctx._chart.destroy();
153
+ ctx._chart = new Chart(ctx, {
154
+ type: 'doughnut',
155
+ data: {
156
+ labels,
157
+ datasets: [{ data, backgroundColor: colors, borderWidth: 0, hoverOffset: 4 }]
158
+ },
159
+ options: {
160
+ responsive: true,
161
+ maintainAspectRatio: false,
162
+ cutout: '72%',
163
+ plugins: {
164
+ legend: { position: 'bottom', labels: { color: '#7fa8c9', font: { family: 'Share Tech Mono', size: 10 }, padding: 12 } }
165
+ }
166
+ }
167
+ });
168
+ return ctx._chart;
169
+ }
170
+
171
+ // ── Sidebar HTML ──────────────────────────────────────────
172
+ function renderSidebar(activePage) {
173
+ return `
174
+ <div class="sidebar">
175
+ <div class="sidebar-logo">
176
+ <div class="logo-text">SENTINEL</div>
177
+ <div class="logo-sub">FRAUD DETECTION SYSTEM</div>
178
+ <div class="status-pill"><span class="status-dot"></span>SYSTEM ONLINE</div>
179
+ </div>
180
+ <div class="nav-section">
181
+ <div class="nav-label">NAVIGATION</div>
182
+ <a href="/" class="nav-item ${activePage==='home'?'active':''}"><span class="nav-icon">⊞</span>Overview</a>
183
+ <a href="/monitor" class="nav-item ${activePage==='monitor'?'active':''}"><span class="nav-icon">β—‰</span>Live Monitor</a>
184
+ <a href="/analyzer" class="nav-item ${activePage==='analyzer'?'active':''}"><span class="nav-icon">⬑</span>Analyzer</a>
185
+ <a href="/performance" class="nav-item ${activePage==='performance'?'active':''}"><span class="nav-icon">β–²</span>Performance</a>
186
+ <a href="/shap" class="nav-item ${activePage==='shap'?'active':''}"><span class="nav-icon">β—ˆ</span>SHAP Explain</a>
187
+ </div>
188
+ <div class="sidebar-footer">
189
+ <div class="api-status" id="apiStatusText">● CHECKING API...</div>
190
+ </div>
191
+ </div>`;
192
+ }
193
+
194
+ function renderTopbar(title) {
195
+ return `
196
+ <div class="topbar">
197
+ <div class="page-title">${title}</div>
198
+ <div class="topbar-right">
199
+ <div class="live-badge"><span class="live-dot"></span>LIVE</div>
200
+ <div class="clock" id="clock">--:--:-- IST</div>
201
+ </div>
202
+ </div>`;
203
+ }
204
+
205
+ async function checkApiStatus() {
206
+ const data = await apiFetch('/health');
207
+ const el = document.getElementById('apiStatusText');
208
+ if (el) {
209
+ if (data && data.status === 'online') {
210
+ el.innerHTML = '<span style="color:#00ff88">● API ONLINE</span>';
211
+ } else {
212
+ el.innerHTML = '<span style="color:#ff2d55">● API OFFLINE</span>';
213
+ }
214
+ }
215
+ }
216
+
217
+ document.addEventListener('DOMContentLoaded', () => {
218
+ const clockEl = document.getElementById('clock');
219
+ if (clockEl) startClock(clockEl);
220
+ checkApiStatus();
221
+ });