Banking Fraud Detection Using Optimized Enhanced Stacked Autoencoder Approach

ABSTRACT Recently, different fraudulent activities have been reported across different fields like finance, insurance, and banking. Banking fraud is the unauthorized exploitation of credit card information to gain access to funds and purchase products. It involves the manipulation of card details for fake transactions. These fraudulent activities lead to significant financial losses for businesses across various industries. This work presents a model for credit card fraud detection using a hybrid approach that merges stacked autoencoder (SAE) and long short term memory (LSTM) networks and hyper‐parameter optimization by the walrus optimization algorithm (WOA). The SAE is presented to capture automated features and minimize the data dimensionality. The LSTM network is utilized for the identification of fraudulent transactions The WOA is considered for optimizing the hyperparameters of the model and preventing overfitting issues. The proposed SAE‐LSTM‐WOA model is evaluated on three benchmark datasets and attained better accuracies of 99.05% (European Cardholders), 99.03% (German credit data), and 99.17% (Inclusive Indian Fraud Dataset). Thus, the outcomes show the ability of the model to generalize across various data and offer improved reliability in credit card fraud detection cases.