jemimahkaberia/mpesa_fraud_detection

M-Pesa mobile money fraud detection using XGBoost and SMOTE — ROC-AUC 0.9990 --- title: M-Pesa Fraud Detection emoji: 🔍 colorFrom: teal colorTo: green sdk: gradio sdk_version: 4.0.0 app_file: app.py pinned: true --- ## M-Pesa Fraud Detection Mobile money fraud is a real problem in Kenya. I built this to show how machine learning can catch it. Trained on 6.3 million transactions from the PaySim dataset. The model looks at things like whether a sender's account was completely drained, whether the recipient's balance didn't change after receiving money, and what fraction of the sender's balance was transferred — patterns that show up consistently in fraudulent transactions. ## How it works Enter the transaction details and click Check Transaction. It tells you instantly whether the transaction looks suspicious and why. ## Results | Model | ROC-AUC | Fraud Precision | Fraud Recall | |---|---|---|---| | Logistic Regression | 0.9951 | 13% | 95% | | Random Forest | 0.9990 | 97% | 100% | | XGBoost | 0.9987 | 96% | 100% | Random Forest came out on top — it caught every single fraud case with 97% precision. ## Tools used Python, Pandas, Scikit-learn, XGBoost, SMOTE, Gradio ## About Built by Jemimah Mugure Kaberia BSc Data Science, Co-operative University of Kenya github.com