FRAUDDETECTNET: A CAPSULE NETWORK WITH CNN-LSTM HYBRID MODEL FOR REAL-TIME BANKING FRAUD

With the dynamic digital banking ecosystem, fraud detection is still an important issue. This paper introducesFraudDetectNet, a hybrid Capsule Network with CNN-LSTM model, which can detect real-time bank fraud. Themodel integrates CNN's spatial feature extraction power, LSTM's learning of temporal dependencies, and thehierarchical representation strength of Capsule Networks. Applying the PaySim data, the envisioned modelexhibits robust performance with an accuracy of 99.38%, precision rate of 99.42%, recall rate of 99.32%, and F1-score of 99.37%. The system gains a False Positive Rate of 0.565% and a False Negative Rate of 0.675% and thuswill be extremely accurate for real-time fraud detection across cloud-based bank systems