Abstract : This paper presents a hybrid ARIMA–machine learning (ARIMA–ML) regression framework designed to improve predictive accuracy in cyber‑physical systems (CPS). The approach brings together the strengths of classical statistical time‑series modelling and modern data‑driven techniques, allowing the model to capture both linear structures and nonlinear dynamics that commonly arise in CPS environments. A simulation‑based evaluation was conducted using a multivariate dataset generated from a MATLAB/Simulink CPS model, complemented by Python‑based machine learning components. The results show that the hybrid model consistently outperforms standalone ARIMA and ML approaches across multiple operational scenarios, including normal operation, peak load, and early‑stage failure conditions. Improvements were observed not only in RMSE and MAE but also in residual stability, prediction interval reliability, and statistical significance as confirmed by the Diebold–Mariano test. These findings suggest that hybrid modelling offers a practical and effective pathway for enhancing predictive maintenance, anomaly detection, and decision‑support capabilities in complex CPS environments. Future work will explore real‑time deployment, integration with edge computing platforms, and the use of more advanced learning architectures to further strengthen model adaptability and performance.
Ass. Eng. Iliyan Vasilev,, PhD (Mon,) studied this question.