Optimizing E-Commerce Fraud Detection with BiGRU and Capsule Network Architectures

Computers and enterprises have proliferated to the point that most financial transactions may now be conducted through electronic commerce systems. This includes systems for credit, telephones, healthcare insurance, etc. The truth is that these solutions are used by both law-abiding citizens and criminals. Con artists also tried a variety of techniques to break into the e-commerce platforms. To provide adequate security for the e-commerce platforms, fraud prevention systems (FPSs) disappoint. Preprocessing, feature selection, and training the model are the three stages that make up the suggested method. Preprocessing includes discretization and min max normalization. Discretization is used to shorten attribute intervals and normalization is used to divide attribute values. Improving the efficacy of machine learning (ML) models in the intrusion detection system domain by GA-based feature selection. When training the model, a BiGRU-A-CapsNet was utilized. The suggested method outperforms BiGRU and CapsN et with an average accuracy of 95.44 percent.