Predictive modeling for design of low elastic modulus and biocompatible β-Ti alloys: A comprehensive machine learning-data driven approach with XGBoost as the key predictor

Abstract

Mechanical compatibility and biocompatibility are critical considerations in the design of β-Ti alloys. The former hinges on achieving a reduced elastic modulus, while the latter is contingent on the toxic alloying element’s absence. Although conventional alloy design methods have been employed for developing low-modulus β-Ti alloys, these theoretical approaches encounter limitations due to the high non-linear relationship between composition and elastic modulus. To address these challenges and improve the design of low-modulus and biocompatible β-Ti alloys, an alternative approach utilizing machine learning (ML) techniques has been explored. ML leverages sophisticated algorithms to discern intricate connections between material composition and elastic modulus, making it a promising avenue for alloy design. In this study, seven supervised ML regression models were developed and compared to identify the most effective model for the given dataset. Notably, the XGBoost model exhibited superior performance compared to the other models, boasting a remarkable accuracy with an R-squared (R²) value of 0.962 and an RMSE (Root Mean Square Error) of 3.16 GPa for the test data. Due to the inherent complexity of comprehending the structures and patterns present within datasets and simultaneously formulating effective predictive models, this study pursued the exploration of dependable models through the avenue of statistical inference analysis, a novel approach yet unexplored in prior research studies. Furthermore, a feature importance analysis was conducted to ascertain crucial input parameters significantly influencing the output, based on the best-performing model. The validity of the best model’s results was confirmed by comparing them with experimental data, ensuring the reliability and applicability of the ML-based alloy design method.