Machine learning-guided design of aluminum–silicon alloys with balanced strength and ductility

ABSTRACT Al-Si alloys, valued for their superior mechanical properties, are widely deployed in aerospace, automotive, and electronics applications. A persistent challenge in their design lies in the intrinsic trade-off between tensile strength (TS) and elongation (EL), which often compromises balanced mechanical performance. To address this limitation, we developed rigorous 5-fold cross-validated ensemble learning models. The ensemble model achieves high accuracy for TS prediction (R 2 = 0.9308) and moderate but acceptable accuracy for EL prediction (R 2 = 0.7410), demonstrating robust predictive capability—particularly for strength parameters. One optimized sample with composition of Al-6.63Si-3.53Cu-0.67Ni-0.25Fe-0.22Mn was experimentally characterized, achieving a TS of 315.78 MPa and EL of 9.33%. These results generally consistent model predictions, helping mitigate the conventional strength-ductility trade-off. In addition to its high accuracy in predicting mechanical properties, the model achieves good interpretability, as feature importance analysis identifies the key compositional and processing factors determining mechanical performance.