Hybrid Machine Learning-Based CPU Performance Prediction for Heterogeneous Scheduling Using LASSO, SVR, and Metaheuristic Optimization

Abstract

The increasing prevalence of heterogeneous computing systems necessitates the development of advanced CPU scheduling strategies capable of leveraging diverse computational resources effectively. Accurate performance prediction of applications across heterogeneous platforms is a key enabler of efficient scheduling. In this study, machine learning (ML) models—Lasso Regression (LAS) and Support Vector Regression (SVR)—were employed to predict CPU performance. To enhance their accuracy, two metaheuristic optimization algorithms, the Crocodiles Hunting Strategy (CHS) and the Walrus Optimization Algorithm (WOA), were integrated with the base models, resulting in four hybrid schemes: SVCH (SVR + CHS), SVWO (SVR + WOA), LACH (LAS + CHS), and LAWO (LAS + WOA). The models were trained and validated on a dataset comprising 205 samples from published papers. Model performance was evaluated using five metrics: Root Mean Square Error (RMSE), Coefficient of Determination (R²), Mean Squared Error (MSE), Symmetric Mean Absolute Percentage Error (SMAPE), and Coefficient of Variation of RMSE (CV_RMSE). Among all models, the LACH scheme achieved the best validation performance with an R² of 0.993, RMSE of 8.778, and MSE of 77.1, significantly outperforming the base SVR model, which achieved an R² of only 0.915 and RMSE of 32.785. Compared to existing ML-based CPU prediction approaches, the proposed hybrid models—particularly LACH and LAWO—demonstrate superior accuracy and robustness, offering valuable potential for enhancing scheduling decisions in heterogeneous systems.