This study presents a robust and novel framework by integrating gradient boosting machine (GBM) models with three nature-inspired metaheuristic algorithms, namely, gorilla-troop optimization (GTO), grey wolf optimization (GWO), and biogeography-based optimization (BBO), to predict the swell pressure ( \(\:{P}_{s}\) ) of expansive soils. Notably, very few studies have utilized hybrid ensemble soft-computing models to evaluate the \(\:{P}_{s}\) of expansive soils. The models were trained on 168 datasets comprising nine key geotechnical parameters. All the hybrid models are concluded to outperform the traditional standalone model, outlining the importance of hybridization. Among the hybrid models, GBM-GWO outperformed others, with R² = 0.9570 (training) and 0.9397 (testing), RMSE = 0.0404 and 0.0169, and MAE = 0.0297 and 0.0090, respectively. Residual error and comprehensive performance analysis confirmed its superior accuracy and minimal overfitting. Global sensitivity analysis identified specific gravity (Si = 0.394), maximum dry density (Si = 0.391), and swelling potential (Si = 0.378) as the most influential parameters. The proposed models offer a reliable, soft-computing, data-driven approach for the efficient assessment of swelling pressure.
