Understanding the link between material composition and mechanical properties is crucial for material design and optimization. This study utilized four machine learning models—K-Nearest Neighbor (KNN), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN)—to predict tensile strength and examine how alloying composition affects the tensile strength of titanium (Ti) alloys. The first three models are considered simpler machine learning approaches, while the ANN is a deep learning method. The models were trained on publicly available experimental data, using fifteen alloying elements as inputs: Al, C, Cr, Cu, H, Fe, Mo, Ni, Nb, N, O, Si, Sn, V, and Zr. A new framework was proposed for evaluating the best model, which involves running several ML models, assessing metrics such as R2 value, absolute percentage error distribution, and stability of R2 through multiple trials, and finally comparing model accuracy using the Diebold-Mariano test. The evaluation showed that all four models achieved good accuracy, with R2 values above 80%. However, the framework identified KNN as the best model due to its low error rate, the narrowest range in absolute percentage error, and more stable R2 value. Additionally, the feature importance analysis highlighted how alloying elements impact tensile strength, revealing both linear and non-linear correlations. It was found that increasing Ti content and using alloying elements with an atomic radius smaller than Ti affect the tensile strength. This study illustrates the potential of machine learning in material screening and design for Ti alloys.
