Machine learning-assisted prediction of swelling behavior in sodium alginate/triethylene glycol/acrylic acid pH-responsive biodegradable hydrogels

pH-responsive hydrogels have attracted considerable attention for biomedical applications because of their ability to undergo controlled swelling under varying environmental conditions. However, accurately predicting swelling behavior remains challenging due to the complex nonlinear interactions between pH and swelling kinetics. In this study, sodium alginate/triethylene glycol/acrylic acid (STA) biodegradable hydrogels were investigated, and an optimized machine learning framework was developed to predict their swelling behavior under different pH conditions. Experimental swelling data collected over 10–360 min at pH 4, 6, 7.4, 8, and 10 were modeled using an optimized Gradient Boosting regression algorithm with engineered physicochemical features and randomized hyperparameter optimization under 10-fold cross-validation. The optimized model achieved excellent predictive performance with a coefficient of determination (R2) of 0.9617, a root mean square error (RMSE) of 202.22, and a mean absolute error (MAE) of 116.36, demonstrating strong agreement between experimental and predicted swelling values. Feature importance and SHapley Additive exPlanations (SHAP) analyses identified nonlinear time-dependent descriptors and pH–time interaction features as the dominant factors governing swelling prediction, improving model interpretability. The proposed framework provides an accurate, interpretable, and computationally efficient tool for hydrogel swelling prediction, supporting the rational design and optimization of pH-responsive biomaterials.