Enhancing Real Time-Fraud Detection in Banking Transactions using Recurrent Neural Networks (RNNs)
Ashok Ghimire, Sandeep Shrestha, Paavani Jain, Reshma Leslie · International Journal of Innovative Research in Science Engineering and Technology · 2025
The transformation from traditional methods to the so-called Internet-based banking services has resulted in a huge surge in attack possibilities for fraudulent instantiations in the financial transactions. Traditional fraud detection mechanisms controlled by set rules may not be able to keep up with emerging patterns of fraud, resulting in either late countermeasures or irreversible financial losses. This paper discusses the application of RNN for increasing the accuracy of real-time fraud detection for banking transactions. Recurrent neural networks, being capable of working with sequential data and learning temporal dependencies, can uncover very subtle anomalies within transaction sequences indicative of fraudulent behavior. The application model is trained on historical transaction data in order to capture normal customer behavior patterns and identify deviations in real-time. LSTM and GRU structures are used to overcome vanishing gradient problems and to obtain better, more accurate predictions. The obtained experimental results confirm that RNN-based models can outperform classical machine learning methods concerning detection accuracy, false positive rate, and real-time processing. This work suggests that RNNs can further improve the safety and reliability of financial systems by providing more adaptive and responsive fraud detection mechanisms.