Accurate prediction of surface roughness (SR) and material removal rate (MRR) in Wire Electrical Discharge Machining (WEDM) of Al7075/Al2O3 composites remains a critical challenge due to complex, nonlinear interactions among machining parameters and the lack of robust predictive models. Additionally, achieving reliable SR prediction is difficult because of the stochastic nature of the process. In this study, Al7075/Al2O3 composite was fabricated using the stir casting process to ensure uniform particle distribution. The WEDM experiments were systematically designed using Taguchi’s L₁̄8 orthogonal array, where voltage, current, pulse-on time, pulse-off time, and bed speed were varied to evaluate their effects on SR and MRR. This study addresses these issues by developing predictive models using CatBoost, Decision Tree, and Naïve Bayes algorithms. Experiments were designed using Taguchi’s L₁̄8 orthogonal array by varying key parameters. When it comes to predicting MRR and SR, the CatBoost algorithm and the Decision tree method outperform naive Bayes. CatBoost achieved higher prediction accuracy (0.83), precision (1.0), and ROC-AUC (0.83) for MRR compared to other models, indicating its effectiveness in capturing nonlinear relationships without overfitting. Satisfactory accuracy of predictions is obtained using decision tree models, whereas Naïve Bayes gives the least accuracy because of the conditional independence of assumptions made by the algorithm. The lower prediction accuracy for surface roughness (maximum accuracy of 0.5 and ROC-AUC up to 0.5) suggests higher complexity in its prediction, without attributing it solely to randomness or proposing unverified solutions that can be solved by hybrid machine learning models in future. Results show CatBoost outperforms others for MRR prediction, while SR prediction remains less accurate, highlighting the need for advanced hybrid modeling approaches.
Arunadevi et al. (2026) studied this question.