Machine learning-based prediction of volumetric wear rate in hybrid ZA-27 composites

Abstract Accurate prediction of wear performance in composite materials is essential for reducing experimental costs and improving material design in tribology. ZA-27 alloy-based hybrid composites reinforced with SiC and graphite particles exhibit enhanced wear resistance but require labor-intensive testing for performance evaluation. This study develops a predictive modeling framework by integrating synthetic minority over-sampling technique for regression (SMOTER) with machine learning (ML) algorithms to estimate volumetric wear rate. The dataset consisted of four input variables (SiC ratio, graphite ratio, load, sliding speed) and one output variable (volumetric wear rate). SMOTER was applied to address the imbalanced distribution of wear data. Linear regression (LR), support vector regression (SVR), decision tree regression (DTR), artificial neural networks (ANN), and adaptive neuro-fuzzy inference system (ANFIS) were evaluated. ANFIS achieved the highest performance with R 2 = 0.9291, MSE = 0.0428, MAE = 0.0483, MAPE = 0.401, KGE = 0.9361, and NSE = 0.9186. SVR and ANN also demonstrated robust performance, with R 2 values of 0.8935 and 0.8941, and MAPE values of 0.363 and 0.371. In contrast, LR and DTR yielded considerably lower predictive performance. The study demonstrates that ML-based approaches can effectively model tribological behavior, reduce experimental burden, and support the development of advanced composite materials.