Prediction of mechanical properties of biomedical magnesium alloys based on ensemble machine learning

In this work, a dataset was constructed by collecting the data of mechanical properties for 365 magnesium (Mg) alloys. Using the composition and process parameters of Mg alloys as input variables, six machine learning (ML) models including ridge regression, support vector machine regression, gradient boosting regression tree, random forest, CatBoost, and Gaussian process regression, were built to predict the ultimate tensile strength (UTS), yield strength (YS), and elongation (EL) of Mg alloys. These single models were then integrated by using model ensemble in order to further improve the prediction accuracy. The results showed that the ensemble model achieved a higher prediction accuracy and better generalization ability for UTS, YS, and EL than that for the single models. The mechanical properties predicted by the optimal model were very close to the experimental values, demonstrating that ML is an effective method for predicting the mechanical properties of biomedical Mg alloys.