This study developed machine learning (ML) models to predict the mechanical properties of Ni-free β-type titanium shape memory alloys (SMAs). Using a dataset of 107 entries derived from both literature and laboratory experiments, we focused on predicting ultimate tensile strength (UTS) and elongation (EL). Key features, including Mo equivalent, bond order, and d-orbital energy level, were selected for the models through Pearson correlation maps and subset selection methods. Four ML algorithms—Linear Regression (LIN), Support Vector Regression (SVR), Random Forest Regression (RFR), and Gradient Boosting Regression (GBR)—were employed and evaluated using metrics like mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R2). The GBR model for EL showed the highest prediction accuracy (R2 = 0.998 for training and R2 = 0.817 for testing), whereas UTS predictions were less accurate (R2 < 0.6 for testing). Although the models were also adapted to predict yield stress (YS), their accuracy was reduced, with improvements seen when incorporating phase constitution information reflecting phase stability. The primary reasons for the discrepancy in this study include the small dataset size and the absence of microstructural features. This research demonstrates the potential of ML models in predicting the mechanical properties of β-type titanium SMAs, highlighting the importance of integrating domain-specific knowledge through feature engineering to overcome the challenge of small data sets, and to enhance accuracy and robustness.