<![CDATA[Cellulose-based hydrogels hold great promise for biomedical, agricultural, and materials engineering applications due to their exceptional water absorption capacity. However, accurately predicting swelling degree remains challenging due to the complex nonlinear relationships between synthesis parameters. This study presents a machine learning approach to predict the swelling degree of cellulose hydrogels using an experimentally compiled dataset from the literature, comprising 124 samples with 11 input features encompassing NaOH and urea concentrations, cellulose type, crosslinker type and concentration, and process parameters. Fourteen machine learning algorithms were trained and compared, including Extra Trees, Gradient Boosting, XGBoost, LightGBM, Random Forest, Decision Tree, SVR, KNN, Linear Regression, Ridge, Lasso, ElasticNet, and ANN (MLP). Preprocessing involved KNN imputation of missing values, logarithmic transformation of the target variable, and interaction feature engineering. Outlier detection was performed using three methods: IQR, Z-score, and Isolation Forest. Results show that the Extra Trees model achieved the best performance with R² = 0.9684, RMSE = 0.3382, and MAE = 0.2497 on the logarithmic scale. After hyperparameter tuning, the model achieved R² = 0.9669 and RMSE = 0.3462 on the log scale, and R² = 0.9574, MAE = 488.19%, and RMSE = 1291.60% on the original scale. SHAP analysis identified crosslinker type and cellulose group as the most dominant features, followed by soaking time and temperature, and cellulose percentageThis approach provides an efficient, data-driven quantitative framework for optimizing hydrogel formulations.]]>
