A new system detects anomalies in financial transactions by combining three different models and improving detection accuracy.
Researchers created a system to detect anomalies in financial transactions by combining three models: Isolation Forest, LSTM Autoencoder, and One-Class SVM. The system is able to detect subtle deviations in operational metrics and silent failures that occur without explicit exceptions. Experimental evaluation showed that the system improves detection robustness and reduces false negatives compared to individual models and simple averaging strategies.
Abstract
Mobile payment systems process millions of transactions daily across distributed microservice architectures, where operational anomalies and silent failures can lead to financial losses and system instability. Traditional threshold-based monitoring is insufficient for detecting subtle, context-dependent deviations that evolve with user behavior and workload patterns. This study introduces a self-learning hybrid anomaly detection framework that integrates Isolation Forest, LSTM Autoencoder, and One-Class SVM to capture statistical, temporal, and structural deviations in operational metrics. Model outputs are fused using a calibrated soft majority voting strategy based on normalized anomaly scores. The trained framework is deployed as a containerized microservice, enabling real-time anomaly assessment based on live operational statistics. Experimental evaluation across a fifteen-month dataset demonstrates that the ensemble improves detection robustness and reduces false negatives compared to individual models and simple averaging strategies. The results highlight the system’s ability to detect silent failures and abnormal behaviors that occur without explicit exceptions while maintaining scalability and adaptability in complex financial microservice environments.
References
- 1.Dragoni, N., Giallorenzo, S., Lafuente, A. L., Mazzara, M., Montesi, F., Mustafin, R., & Safina, L. (2017). Microservices: yesterday, today, and tomorrow. Present and ulterior software engineering, 195-216.
- 2.Liu, F. T., Ting, K. M., & Zhou, Z. H. (2008, December). Isolation forest. In 2008 8th IEEE International Conference On Data Mining (pp. 413-422). IEEE.
- 3.Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly detection: A survey. ACM computing surveys (CSUR), 41(3), 1-58.
- 4.Campos, G. O., Zimek, A., Sander, J., Campello, R. J., Micenková, B., Schubert, E., ... & Houle, M. E. (2016). On the evaluation of unsupervised outlier detection: measures, datasets, and an empirical study. Data mining and knowledge discovery, 30(4), 891-927.
- 5.Pankaj, M. (2016). LSTM-based encoder-decoder for multi-sensor anomaly detection. CoRR, 1607.
- 6.Schölkopf, B., Platt, J. C., Shawe-Taylor, J., Smola, A. J., & Williamson, R. C. (2001). Estimating the support of a high-dimensional distribution. Neural computation, 13(7), 1443-1471.