Enhancing Financial Fraud Detection Through Chimp-Optimized Long Short-Term Memory Networks

Abstract

The proliferation of online shopping has led to a substantial increase in payment card transactions, accompanied by a parallel rise in fraudulent activities. Such frauds impose significant financial burdens on both businesses and banking institutions annually. In response to this growing concern, a novel hybrid methodology has been developed, integrating a metaheuristic optimization algorithm with a neural network classifier, aimed at the automatic detection of financial transaction fraud. This method, termed ChimpOptimized Long Short-Term Memory Networks (ChOpt+LSTM), operates in two sequential phases. Initially, an optimization algorithm based on chimp behavior is utilized for the selection of the most pertinent features for fraud detection. Subsequently, these features inform the training of a Long Short-Term Memory (LSTM) classifier model, specifically designed for the identification of credit card fraud. An extensive comparative analysis reveals that the proposed ChOpt+LSTM method surpasses existing techniques in several key performance metrics. Notably, it achieves a classification accuracy of 99.18%, a mean absolute error (MAE) reduction to 25.7, a mean squared error (MSE) reduction to 16.3, alongside precision, recall, and F1 scores of 98.54%, 98.47%, and 96.58%, respectively. These findings underscore the efficacy of combining chimp optimization algorithms with LSTM classifiers in enhancing the accuracy and reliability of financial fraud detection systems.