Decoding Processing Effects in Al Alloys: A Data-Driven Approach Using Explainable Artificial Intelligence

Understanding the complex relationships between processing conditions and mechanical properties in aluminum alloys remains a critical challenge in materials science. This study presents a data-driven framework using explainable artificial intelligence to quantify and interpret how different processing routes influence the strength–ductility trade-off in aluminum alloys. Using a comprehensive dataset of 1154 aluminum alloy samples with 10 distinct processing conditions, optimized XGBoost models were developed via Bayesian hyperparameter tuning to predict yield strength (R2 = 0.9392), tensile strength (R2 = 0.9491), and elongation (R2 = 0.6767). The strength models showed high predictive accuracy, whereas elongation showed lower and less uniform reliability, with the largest relative errors in the 0–5% elongation regime. SHAP (SHapley Additive exPlanations) analysis revealed that processing condition is the most influential feature for yield strength prediction, while Cu dominates tensile strength prediction. True SHAP interaction analysis identified Processing_encoded interactions with Cu as the strongest processing-coupled contribution, followed by Mg and Al, with Zn, Si, and Li showing smaller but non-negligible interaction contributions. The decision-tree surrogate is presented as an exploratory rule-extraction tool rather than as a standalone processing-selection classifier. These findings demonstrate that explainable Machine Learning (ML) can support interpretation of processing–property relationships in aluminum alloys when predictive limitations, class imbalance, and the associative nature of SHAP explanations are explicitly considered.