Credit Card Fraud Detection

The project delves into the challenging domain of credit card fraud detection, a critical task for financial security. Using a real-world (but highly imbalanced dataset) of anonymized European transactions, the goal was to develop a robust model to identify fraudulent activity.

Initially, a standard approach yielded an impressive 99.93% accuracy, but this was misleading due to the severe data imbalance (only 0.172% of transactions were fraudulent). The model was simply performing well by correctly identifying the vast majority of legitimate transactions, while often missing the fraudulent ones.

To address this, the project employed a strategic data resampling technique, balancing the dataset to improve the model’s ability to learn the subtle patterns of fraud. By applying an optimized XGBoost model to this balanced data, I achieved a more meaningful performance with an accuracy of 93%, indicating a significant improvement in detecting the minority fraud class.