Predicting the Young's Modulus of biomedical titanium alloys using machine learning: a data-driven approach.
TL;DR
Comparative analysis with XGBoost, Random Forest, Gradient Boosting, and Support Vector Machine confirmed the capability of the proposed DNN model for capturing complex behavior non-linear relationships in Ti-Alloys data.
Abstract
This study presents a machine learning framework for predicting the Young's Modulus (YM) of biomedical titanium alloys to address stress shielding in implant applications. A Deep Neural Network (DNN) was developed using nineteen compositional features and physically meaningful descriptors. Prior to model training, the dataset was preprocessed using Box-Cox and Yeo-Johnson transformations to improve data distribution while preserving all samples. The model architecture incorporates multiple hidden layers with L1/L2 regularization and dropout to enhance generalization. Training was conducted using early stopping, terminating at 585 epochs to prevent overfitting. The optimized model achieved a testing Mean Squared Error (MSE) of 0.294 Gpa and r2-score of 0.82, demonstrating predictive performance. Comparative analysis with XGBoost, Random Forest, Gradient Boosting, and Support Vector Machine confirmed the capability of the proposed DNN model for capturing complex behavior non-linear relationships in Ti-Alloys data.