Publication

Abstract : The patterns of fraud across transactions have never been more complex, creating a challenge to financial institutions in terms of assessing and reducing risk. An integrated hybrid Quantum Machine Learning (QML) framework, TransGuard-QNet, is presented in this research for financial transaction risk assessment and prediction, which combines supervised and unsupervised models with Linear Programming (LP) for decision optimization. After preprocessing the input features with several preprocessing techniques, the proposed model oversees outliers via Isolation Forest (iForest) for improved robustness. A feature engineering phase primarily relies on mathematics, behaviors, statistics, frequency, and correlation features. Besides, it uses an Improved Extreme Gradient Boosting (iXGBoost) and Quantum Neural Nets (QNNs) model to predict risk probability, and Quantum Autoencoders (QAEs) and k-Means clustering techniques to find anomalies. The resulting predictions are combined to form a final output based on the optimized parameters with a focus on achieving both high predictive accuracy and anomaly detection effectiveness. Here, the proposed LP is used to determine the weights of the proposed TransGuard-QNet. With this methodology, expected financial losses are reduced, and resource usage is optimized. Model performance is measured by several key metrics: accuracy (99.09%), precision (99.51%), recall (98.51%), F1-score (98.59%), and the economic impact. This approach has also outperformed existing methods with significant gains in detecting fraud as well as in operational efficiency and resource utilization. Thus, making it a promising candidate for deployment in dynamic transaction environments to manage financial risk.