A machine learning project for identifying potentially fraudulent transactions in highly imbalanced transaction data.
Credit card fraud detection is a classification problem on highly imbalanced data — very few fraudulent transactions among millions of legitimate ones. High recall is essential to catch fraud, but precision matters too because false positives create friction for genuine customers.
These are reported model evaluation results. Fraud detection is not perfect, and these metrics do not imply every fraudulent transaction will be caught.
The precision-vs-recall trade-off is central here. High recall means fewer missed frauds, but if precision drops the business faces more false positives. SHAP explainability helps interpret which features drive predictions and supports more transparent review of flagged cases.
No live deployment link exists for this project.