A Hybrid Deep Learning Model For Detecting Credit Card Fraud Using Cnn–bilstm With An Attention Mechanism And Focal Loss Optimization

Abstract

Detecting Credit Card Fraud Is A Complex Task Due To The Significant Imbalance In Data And The Constantly Changing Nature Of Fraudulent Activities. This Research Introduces A Hybrid Deep Learning Model That Combines CNN, BiLSTM, And An Attention Mechanism To Effectively Identify Spatial And Temporal Patterns In Transactions. To Maintain Regional Specificity, Separate Models Were Developed For Datasets From India And Europe. To Tackle Class Imbalance Without Introducing Synthetic Bias, Focal Loss With Adaptive Class Weighting Was Employed. The Experiments Revealed That The Model For The European Dataset Achieved An Accuracy Of 99.32% And An F1-score Of 97.29%, While The Model For The Indian Dataset Reached An Accuracy Of 98.67% And An F1-score Of 95.80%. The Use Of Attention Mechanisms Enhanced The Relevance Of Features And Overall Performance, And SHAP-based Explainability Improved The Interpretability Of The Model. The System Was Deployed Using Stream Lit For Real-time Fraud Detection, Offering A Scalable And Precise Solution Tailored To Region-specific Financial Fraud Detection.

Keywords