A Hybrid Framework for Fraud Detection in Mobile Payment System based on RNN and Hidden Markov Model

Any fraudulent misrepresentation for monetary gain constitutes a risk of fraud. With the increasing popularity of mobile payment alternatives, the identification of fraud in digital currency trading platforms is essential. Efficient fraud detection is essential for transaction security and the protection of financial losses. The detection of fraud in mobile payment systems is enhanced through preprocessing, feature selection, and model training. Data preprocessing, particularly data cleansing to mitigate modelling approach sensitivity, is crucial for transforming raw data into significant insights. Information Gain, based on Entropy, identifies the most pertinent attributes. A hybrid RNN-HMM is trained on the chosen features to enhance fraud detection. The proposed RNN-HMM model surpasses the performance of separate RNN and HMM approaches. The 97.38% accuracy rate in the experiments indicates success. This study underscores the necessity for robust mobile payment fraud detection, particularly in the realm of digital currency trading. Advanced DL techniques render the RNN-HMM model an exceptionally precise framework for fraud detection. The findings indicate that efficient preprocessing, feature selection, and model training can enhance the security and reliability of mobile payment systems.