giciriri/mpesa-fraud-detection

# FraudGuard — AI-Powered M-Pesa Fraud Detection System > A real-time intelligent fraud detection platform designed to protect M-Pesa users from mobile money fraud, built with Python, Flask, and a trained Random Forest machine learning model. --- ## Overview The system analyses every M-Pesa transaction in real time using a hybrid detection approach that combines machine learning, rule-based logic, and explainable AI to classify transactions as Low, Medium, or High risk — and takes immediate action before fraud occurs. --- ## Key Features - **Real-time AI fraud scoring** using a trained Random Forest classifier - **3-tier risk classification** — Low, Medium, and High risk - **Explainable AI** — every decision comes with a plain-English explanation - **Email OTP verification** for medium-risk transactions - **Instant block and email alert** for high-risk transactions - **Live risk preview** — see the AI working as you type - **Admin intelligence dashboard** with fraud statistics, charts, and manual override - **Simulation mode** — one-click demo scenarios for presentations - **Secure authentication** with hashed PINs and Flask-Login --- ## System Architecture The system uses a three-layer hybrid detection engine: **Input** → Flask REST API → **Hybrid Detection Engine** → **Decision Engine** → **Action** The Hybrid Detection Engine combines three components: - **ML Scoring Engine** — Random Forest model analyses transaction type, amount, and hour - **Rule-Based Engine** — Expert banking rules applied on top of the ML score - **Risk Score** — Combined output classified as LOW, MEDIUM, or HIGH **Actions taken:** | Risk Level | Action | |------------|--------| | 🟢 LOW | Transaction approved instantly | | 🟡 MEDIUM | OTP email sent — user must verify | | 🔴 HIGH | Transaction blocked + security alert email sent | ## Model Performance The Random Forest model was trained on the PaySim dataset containing 209,715 transactions. | Metric | Score | |----------- …