Dataset and Machine-Learning Analysis for Lap-Shear Fracture Load and High-Performance Conditions in HDPE Friction Stir Spot Welding

This dataset and reproducibility package accompanies the revised manuscript entitled “Machine-Learning Prediction of Lap-Shear Fracture Load and Identification of High-Performance Conditions in HDPE Friction Stir Spot Welding.” The experimental dataset contains 258 observations representing 230 unique combinations of varying tool-geometry and process parameters. Each observation includes three lap-shear fracture-load measurements and their arithmetic mean.This version supersedes the machine-learning materials associated with the original manuscript submission. It includes a corrected target value for Experiment 208, for which the three replicate measurements (1470, 1610, and 1520 N) give an arithmetic mean of 1533.33 N. The revised analysis uses eight varying predictors and condition-based grouping to prevent identical experimental conditions from appearing in both training and test subsets.The accompanying Python code implements group-aware model validation, training-only hyperparameter tuning, repeated group-aware evaluation across 20 splits, grouped permutation sensitivity analysis, SHAP interaction analysis, a group-separated split-conformal uncertainty diagnostic, and model-assisted identification of high-performance conditions restricted to experimentally represented conditions. Linear Regression, Support Vector Regression, Random Forest, Gradient Boosting, artificial neural network (ANN), and XGBoost models are evaluated. The package also includes a data dictionary and README with reproducibility information.The model-assisted analysis should not be interpreted as establishing a validated global optimum. The highest experimentally observed mean fracture load was 3694 N, and the revised analysis identifies a descriptive high-performance region within the experimentally represented domain.