from fastapi import FastAPI from pydantic import BaseModel import pandas as pd import joblib from pathlib import Path from fastapi.responses import HTMLResponse import json app = FastAPI(title="Fraud Detection API") # Absolute path for model PROJECT_DIR = Path(__file__).parent.parent.resolve() MODEL_PATH = PROJECT_DIR / "models" / "best_model.pkl" METRICS_PATH = PROJECT_DIR / "models" / "best_run_metrics.json" model = joblib.load(MODEL_PATH) # --- Input Schema --- class TransactionInput(BaseModel): amount_usd: float fee: float source_currency: str dest_currency: str channel: str home_country: str kyc_tier: str account_age_days: int chargeback_history_count: int new_device: bool location_mismatch: bool ip_risk_score: float device_trust_score: float risk_score_internal: float corridor_risk: float txn_velocity_1h: int txn_velocity_24h: int # --- Helper to build features --- def build_features(data: TransactionInput) -> pd.DataFrame: df = pd.DataFrame([data.dict()]) # Flags df["new_device"] = df["new_device"].astype(int) df["location_mismatch"] = df["location_mismatch"].astype(int) # Derived features df["cross_border"] = (df["home_country"] != "US").astype(int) df["device_risk"] = ((df["new_device"] == 1) & (df["device_trust_score"] < 0.5)).astype(int) df["velocity_ratio"] = df["txn_velocity_1h"] / (df["txn_velocity_24h"] + 1) df["velocity_x_amount"] = df["velocity_ratio"] * df["amount_usd"] df["new_account_flag"] = (df["account_age_days"] < 60).astype(int) df["young_account_velocity"] = df["new_account_flag"] * df["txn_velocity_1h"] df["risk_pile_up"] = ( (df["risk_score_internal"] > 0.75).astype(int) + (df["txn_velocity_1h"] >= 2).astype(int) + df["location_mismatch"] + df["new_device"] + df["device_risk"] ) # Align columns with trained model try: MODEL_COLUMNS = model.named_steps["preprocessor"].transformers_[0][2] + model.named_steps["preprocessor"].transformers_[1][2] except Exception: # fallback if preprocessor not found MODEL_COLUMNS = df.columns.tolist() for col in MODEL_COLUMNS: if col not in df.columns: df[col] = 0 return df # --- Endpoints --- @app.get("/") def home(): return {"status": "API is running"} @app.post("/predict") def predict(data: TransactionInput): df = build_features(data) proba = model.predict_proba(df)[0, 1] prediction = int(proba >= 0.5) return {"fraud_probability": float(proba), "prediction": prediction} @app.get("/model-info") def model_info(): if not METRICS_PATH.exists(): return {"error": "No training metrics found. Run main.py to generate a training report."} with open(METRICS_PATH) as f: return json.load(f) @app.get("/dashboard", response_class=HTMLResponse) def dashboard(): if not METRICS_PATH.exists(): return "
Best model: {report['best_model']} (PR-AUC: {report['best_pr_auc']:.4f})
Last trained: {report['trained_at']}
| Model | PR-AUC | Best Threshold |
|---|