Leakage-controlled pre-sintering property prediction in A7075/B 4 C/MXene functionally graded aluminum matrix composites

Abstract Designing A7075/B 4 C/MXene functionally graded aluminum matrix composites requires process, composition and architecture decisions before costly sintering and mechanical testing are completed. This study evaluates whether descriptors available before sintering can support property-oriented experiment prioritization within a defined powder-metallurgy process window. A traceable dataset of 43 specimen codes was assembled from a Taguchi L27 screening backbone, a concentrated MXene-bearing follow-up branch and one-hour milling checks. Separate regression models were developed for Leeb-derived HBW-equivalent hardness, ultimate compressive strength, specific compressive energy absorption and sintered density. Regularized linear and constrained tree-based learners were evaluated using leakage-controlled nested cross-validation, with preprocessing, feature selection, hyperparameter tuning and model selection confined to the corresponding training partitions. Across the three reported seeds, mean outer-fold R 2 ranged from 0.678 to 0.875 for ultimate compressive strength, 0.480 to 0.861 for specific compressive energy absorption, 0.526 to 0.560 for HBW-equivalent hardness and 0.164 to 0.426 for sintered density. Empirical prediction intervals, residual diagnostics, SHAP-based feature attribution and partial-dependence profiles were used to examine model behavior beyond aggregate accuracy values. These interpretation outputs are treated as model-level associations rather than causal metallurgical evidence. Within the investigated A7075/B 4 C/MXene processing and architecture window, the workflow provides a leakage-aware basis for prioritizing pre-sintering experiments, while transfer to other processing routes or independent fabrication campaigns requires further validation.