Optimized Feature Selection and Enhanced Recurrent Neural Network for Financial Fraud Detection

Financial fraud presents a significant challenge worldwide, impeding the steady growth of financial markets. However, detecting fraud is complicated by an imbalanced dataset, where the number of legitimate transactions fraudulent ones. To address this issue, intelligent financial statement fraud detection solutions have been industrialised to assist stakeholders in making informed decisions. This study proposes a novel approach to fraud detection using a Modified Uni-directional Deep-Recurrent Neural Network (MDRNN) model. The model is augmented by the Artificial Rabbit Optimization Algorithm (AROA) to enhance classification accuracy by selecting relevant features. Additionally, the parameters of the MDRNN model were optimized using the Red Fox Optimization (RFO) model. A comparative analysis between the proposed method and existing approaches was conducted and it proved greater efficiency compared to other techniques. The model accomplished an accuracy of 95.65% and identified 591 fraudulent transactions correctly. The outcome of this study would contribute to the improvement of classification accuracy, reduction of misclassification of credit card transactions, and associated cost, and enhancement of financial transaction security.