kushdab/kenya-fraud-check-2025

A machine learning pipeline for detecting identity-theft patterns specifically within Kenyan digital lending apps. # Kenya Fraud Check 2025 ## Overview This project implements a machine learning pipeline tailored for detecting identity theft and fraud patterns within the Kenyan digital lending ecosystem. It focuses on specific indicators such as: - M-Pesa registration status - Device fingerprint reuse counts - Kenyan National ID validity checks - Application time-series patterns (e.g., midnight surges) - Phone number prefix analysis (Safaricom/Airtel/Telkom ranges) ## Features - **Synthetic Data Generator**: Simulates realistic Kenyan loan application metadata. - **Pre-processing**: Handles Kenyan-specific categorical features. - **Model**: Random Forest Classifier optimized for high-precision fraud detection. ## Setup 1. Install dependencies: ```bash pip install -r requirements.txt ``` 2. Train the model: ```bash python train_model.py ``` ## Project Structure - `train_model.py`: Data generation, feature engineering, and model training. - `models/`: Directory where serialized models are stored.