Machine learning prediction of mechanical properties of metallic materials: data augmentation strategies

Accurate prediction of key mechanical properties of metallic materials is essential to accelerating the development and application of new materials. However, existing studies have been insufficient in systematically verifying the effectiveness of data enhancement strategies and comparing the performance of multiple machine learning models, especially in utilizing enhanced datasets to improve prediction accuracy. This study aims to systematically evaluate the impact of data enhancement strategies on the prediction accuracy of the mechanical properties of metallic materials. Based on the publicly available material property dataset, this study constructs a data processing flow that includes data cleaning, feature engineering, oversampling based on the binning of target variables, and synthetic minority class oversampling techniques to generate the original and augmented datasets. On this basis, five machine learning models were trained and evaluated for their prediction performance on both datasets. The results indicate that the applied data enhancement strategies significantly improve the prediction accuracy of all models.