The current study focuses on utilizing machine learning (ML) models in optimizing and predicting laser-induced surface cold hardening in aluminum alloy 6061-O using a pulsed fiber laser. Three input features were utilized: power density (Pd), frequency (f), and pulse overlap (OV), with surface hardness as the matching aim. A set of ML models, including XGBoost and Adaptive Ensemble, was employed. The dataset was preprocessed for normalization and outlier handling, hyperparameter tuning, and assessment through the grid search and cross-validation, and model performance was evaluated using the coefficient of determination (R²) and mean square error (MSE) metrics. The Linear regression and Random Forest regression models were excluded due to their weakness in performance, exhibiting noticeable enhancement in performance of the ensemble. After excluding weak models, the results of XGBoost and Adaptive Ensemble models demonstrated great results of R2/MSE of 0.970/0.774 and 0.949/1.319, respectively. The XGBoost model occupied the first order, followed by the Adaptive Ensemble model, in improving the predictive accuracy to around 96% compared to other models.
