This project aims to detect fraudulent financial transactions using both supervised machine learning models and unsupervised anomaly detection techniques. We used the PaySim synthetic dataset, which simulates mobile money transactions, and applied models like Random Forest, XGBoost, and Isolation Forest to identify suspicious activity.
The project explores class imbalance, feature importance, model evaluation metrics (recall, precision, F1 score, ROC AUC), and compares performance between traditional classifiers and anomaly-based approaches. The end goal is to build a reliable system that can detect fraud accurately while minimizing false positives.