Property Prediction of Medical Magnesium Alloy based on Machine Learning

Abstract

In the research on the performance of magnesium alloy materials as medical implant materials, which is generally measured by plenty of mechanical experiments and corrosion resistance experiments, and this process is time-consuming and inefficient. Machine learning algorithms can train the model based on existing sample data and apply the trained model to predict the data. This paper regards mass fraction of alloying elements, casting or deformation process, heat treatment process and temperature of alloy performance test as features and considers ultimate tensile strength, yield strength, elongation and corrosion rate as labels. It utilizes machine learning to establish least absolute shrinkage and selection operator regression, random forest regression and support vector regression prediction models. The mean absolute error, the mean square error and the coefficient of determination three model evaluation indicators compare the prediction effects of the three regression models. After comparative analysis, the coefficient of determination of the random forest regression model is between 0.8 and 0.9, owning the best prediction effect. The random forest regression model can predict the mechanical properties and corrosion resistance of medical magnesium alloy materials more accurately.