Spaces:

DuckerMaster
/

Fraud-Detection-Multi-Model

Sleeping

App Files Files Community

Fraud-Detection-Multi-Model / README.md

DuckerMaster

Update README.md

bc918a8 verified 5 months ago

preview code

Raw History Blame Contribute Delete

1.28 kB

A newer version of the Streamlit SDK is available: 1.65.0

Upgrade

metadata

title: Fraud Detection Multi-Model
emoji: 🛡️
colorFrom: red
colorTo: blue
sdk: streamlit
sdk_version: 1.57.0
app_file: app.py
pinned: true
suggested_hardware: cpu-basic
python_version: 3.11

Fraud Detection — Multi-Model Live Monitor

Real-time transaction fraud detection using a 3-layer ensemble pipeline.

Architecture

Transaction
    │
    ▼
Layer 1 — Hard Rules
    │  Rule 1: Accessibility + Remote App → Block immediately
    │  Rule 2: Hesitation > 45s + New recipient → bias score ≥ 0.6
    │
    ▼
Layer 2 — Ensemble Scoring
    │  Model A (Sender Behavioral): LightGBM / XGBoost
    │  Model B (Payee Risk Profile): LightGBM
    │  final_score = 0.3 × prob_a + 0.7 × prob_b
    │
    ▼
Layer 3 — Threshold Decision
       score ≥ 0.80  →  Block
       score ≥ 0.50  →  Alert + OTP
       score < 0.50  →  Pass

Files required

File	Purpose
`model_a_pipeline.joblib`	Sender behavioral model
`model_b_pipeline.joblib`	Payee risk profile model
`config.json`	Feature list, weights, thresholds
`payee_profile_store.csv`	Historical payee lookup table

GitHub

Fraud-Detection-Multi-Model