Online Payment Fraud Detection using ML and its Interpretation

Online payment fraud poses a critical threat to the digital financial ecosystem, demanding advanced detection mechanisms to minimize losses. Traditional rule-based approaches often fail to adapt to rapidly evolving fraud strategies, highlighting the potential of machine-learning models. This study evaluates the Random Forest classifier for identifying fraudulent transactions using a dataset of 71,759 records, which are carefully preprocessed to manage class imbalance and remove irrelevant features. The performance of the model was measured using metrics such as precision, recall, and AUC-ROC, with the results showing 98.75% accuracy, 91.07% precision, and 98.72% recall. To improve model transparency, SHAP values were utilized to explain feature contributions. Furthermore, a rule-based chatbot is integrated to assist users. The results affirm the model's effectiveness and reliability in fraud detection while minimizing false positives.