WElvis87/MPESA-Fraud-Detection

# 🛡️ M-PESA Fraud Detection System ## 🎯 Problem Definition: M-PESA is the primary digital payment platform in Kenya, processing millions of peer-to-peer, merchant, and bill payment transactions daily. Its widespread adoption has also made it a target for fraudulent activities, including social engineering scams, account takeovers, SIM-swap fraud, and abnormal transaction behavior. ## 📌 Project Overview M-PESA Fraud Detection is a machine learning initiative designed to identify suspicious transaction patterns. By analyzing transaction metadata, user behavior, and historical fraud markers, the system distinguishes between legitimate financial activity and potential fraudulent attempts. This project demonstrates: - Synthetic and real-world financial data preprocessing - Detection of anomalous transaction patterns - Robust model evaluation for imbalanced datasets The goal is to provide a Decision Support Tool for fintech security teams. By flagging high-risk transactions before they are finalized, stakeholders can mitigate financial loss and enhance user security. ## 📊 Motivation Current fraud detection approaches often rely on rule-based systems and manual reviews, which struggle to adapt to evolving fraud patterns, generate a high number of false positives, and fail to scale with transaction volume. As a result, fraudulent transactions may go undetected, while legitimate users experience unnecessary transaction blocks and delays. The core problem is the lack of a data-driven, adaptive system that can identify potentially fraudulent M-PESA transactions in near real time, while maintaining a low false-positive rate and providing interpretable risk signals for investigation teams. ## 🧠 What This Project Does. The system treats fraud detection as a binary classification problem (Fraud vs. Legitimate). Specifically, it: 1.Analyzes Transaction Flow: Tracks amount, timing, and frequency of transfers. 2.Flags Behavioral Anomalies: Detects sudden spikes in activi …