Real-Time Fraud Detection in Financial Transactions Using Ensemble Methods and LSTM Networks

The objective of this research is to identify the most effective algorithm for real-time fraud detection in financial transactions by assessing the performance of Random Forest (RF), XG Boost, and Long Short-Term Memory (LSTM) models. This research is significant as it aims to enhance fraud detection accuracy, assisting financial institutions in precisely identifying fraudulent activities. Using a dataset obtained from Kaggle, this study examines the predictive capabilities of these algorithms in detecting suspicious transactions. The findings highlight a key observation: LSTM demonstrated superior accuracy in fraud detection compared to Random Forest and XG Boost. This research suggests that machine learning algorithms play a crucial role in improving fraud detection systems. The statistical significance of the results is supported by a p-value of 0.002 (p < 0.05), reinforcing the reliability of the observed differences among the models. Random Forest achieved a mean accuracy of 85.67%, XG Boost yielded 88.01%, and LSTM attained 93.40%. These findings confirm that LSTM outperformed the other models, establishing its effectiveness in real-time fraud detection.