An Optimization-Driven Adaptive Anomaly Detection Framework for Financial and Cyber-Physical Systems Using MATLAB-Based Simulation

Abstract

As the tendency of using more data-driven decision systems in the context of finance and cyber-physical infrastructures is on the rise, the problem of identifying the anomalous and high-risk events has become a critical topic of research. Historical statistical and non-evolutionary machine learning commonly does not adjust to changing data distributions, uncommon cases and game playing actions that typically occur in the real world. In order to overcome these shortcomings, this paper has come up with a demand deceleration adaptive irregularity identification system that incorporates deep learning, evolutionary parameter optimization, and sound decision details. The suggested methodology is tested on MATLAB-based simulated studies on multivariate time-series data of the financial transaction and system activity rates pattern. The architecture is a hybrid between the idea of deep autoencoder based feature learning and adaptive threshold optimization to increase the detection accuracy and resilience in non-stationary settings. The simulation findings reveal that the level of anomaly detection is better than the baseline machine learning models in terms of accuracy, precision, recall and false alarm rate. The paper has emphasized the success of optimization based adaptive learning to light-sensitive risk monitoring systems and has outlined future studies on the application of large scale and real time deployment.

DOI: https://doi-ds.org/doilink/01.2026-72673778