Enhancing Digital Wallet Payments Through Data Analytics: A Study on Fraud Detection and User Personalization.

Abstract

The rapid adoption of digital wallets has revolutionized consumer financial interactions, offering convenience and facilitating the transition to cashless payments. However, this growth has also heightened the risks of fraud, emphasizing the need for advanced detection mechanisms and personalized user services to ensure security and satisfaction. This study presents a hybrid machine learning approach, the Ensemble Anomaly Detection Framework (EADF), which combines Isolation Forest, Local Outlier Factor (LOF), and Long Short-Term Memory (LSTM) networks for enhanced fraud detection. Trained on anonymized transaction data from a leading digital wallet provider, the EADF effectively identifies suspicious activity, achieving a fraud detection accuracy of 97.3% and an F1-score of 0.91, significantly outperforming individual models by an average of 5%. The framework's ability to integrate anomaly detection and deep learning enables precise identification of complex fraud patterns while minimizing false positives. Beyond fraud prevention, the study leverages behavioral data to deliver personalized security prompts and tailored service recommendations, striking a balance between robust security and user convenience. These findings highlight the potential of hybrid models to improve both the safety and user experience of digital wallet platforms. By combining advanced analytics with user-focused design, this research provides a scalable solution for protecting digital transactions and fostering trust in an increasingly cashless economy.