cdoumb/fraud-detection-mobile-money

Mobile Money fraud detection using EDA and Random Forest — PaySim dataset (Africa) # Mobile Money Fraud Detection — PaySim Dataset > End-to-end fraud detection project on Mobile Money transactions — > from exploratory data analysis to a Random Forest classification model. > Based on the **PaySim** synthetic dataset, derived from real transactions > of an **African Mobile Money service**. --- ## Table of Contents - Context - Key Results - Visualizations - Project Structure - Installation - Dataset - Author - License --- ## Context Mobile Money (Wave, Orange Money, Free Money) is a cornerstone of financial inclusion in West Africa. Fraud represents a major risk for these platforms, with fraudsters exploiting TRANSFER and CASH_OUT operations to drain victim accounts. This project analyzes **300,000 transactions** from the PaySim dataset to: - Identify fraud patterns through exploratory data analysis (EDA) - Engineer predictive features based on domain knowledge - Build a Random Forest model to automatically detect fraudulent transactions --- ## Key Results | Metric | Value | |---|---| | Fraud rate in dataset | 0.06% (severe class imbalance) | | Transaction types containing fraud | TRANSFER and CASH_OUT only | | Accounts fully drained in fraud cases | 93.4% | | Model Precision (fraud class) | **100%** — zero false alarms | | Model Recall (fraud class) | **75%** — detects 3 out of 4 fraud cases | | F1-Score (fraud class) | **0.86** | | Most predictive features | Engineered features (`orig_emptied`, `balance_diff_dest`) | --- ## Visualizations ### Transaction Amount Distribution by Type (Fraud vs Legitimate) ### Fraud Count by Transaction Type ### Feature Importance — Random Forest --- ## Project Structure fraud-detection-mobile-money/ │ ├── .gitignore ├── README.md ├── requirements.txt ├── fraud_detection.ipynb # Main notebook (EDA + Feature Engineering + Model) │ └── images/ ├── boxplot_amount_by_type.png ├── fraud_by_type.png └── feature_importance.png --- ## Installation ```bash # 1. Clone the re …