This study presents interpretable machine learning models for predicting glass-forming ability (GFA) in ternary oxide glass systems. Random Forest and XGBoost models were trained using compositional, electronic, thermodynamic, and structural descriptors extracted from the glass_ternary_hipt dataset. The models achieved high predictive accuracy (R² up to 0.929) and identified electronegativity variance and atomic size mismatch as key factors influencing GFA. The results demonstrate the potential of data-driven materials informatics for accelerated glass discovery and optimization.