Low aqueous solubility is a common and serious challenge for most drug substances not only in development but also in the market, and it may cause low absorption and bioavailability as a result. Amorphization is an intermolecular modification strategy to address the issue by breaking the crystal lattice and enhancing the energy state. However, due to the physicochemical properties of the amorphous state, drugs are thermodynamically unstable and tend to recrystallize over time. Glass-forming ability (GFA) is an experimental method to evaluate the forming and stability of glass formed by crystallization tendency. Machine learning (ML) is an emerging technique widely applied in pharmaceutical sciences. In this study, we successfully developed multiple ML models (i.e., random forest (RF), XGBoost, and support vector machine (SVM)) to predict GFA from 171 drug molecules. Two different molecular representation methods (i.e., 2D descriptor and Extended-connectivity fingerprints (ECFP)) were implemented to process the drug molecules. Among all ML algorithms, 2D-RF performed best with the highest accuracy, AUC, and F1 of 0.857, 0.850, and 0.828, respectively, in the testing set. In addition, we conducted a feature importance analysis, and the results mostly agreed with the literature, which demonstrated the interpretability of the model. Most importantly, our study showed great potential for developing amorphous drugs by in silico screening of stable glass formers. Graphical Abstract: [Figure not available: see fulltext.].