Deep Learning‐Based Prediction of Maximum Swelling Pressure in Fine‐Grained Soils

Expansive soils pose significant risks in geotechnical engineering due to their volumetric change behavior. Therefore, accurately predicting maximum swelling pressure is critical for design and safety. In this study, a deep learning‐based model was developed for the prediction of maximum swelling pressure. A dataset of 631 samples, compiled from multiple sources including published literature and additional laboratory measurements, was used. The input variables of the dataset include grain size distribution parameters (sand, silt, and clay percentages), Atterberg limits (liquid limit, plastic limit, and plasticity index), activity, dry unit weight, initial water content, and a derived feature named ClusterLabel. The main contribution of this study is the enhancement of prediction performance by incorporating the ClusterLabel variable, which was generated using the K‐Means clustering method based on maximum swelling pressure values. The proposed deep learning model achieved an average value of 0.774 on the test dataset under five‐fold cross‐validation, outperforming traditional machine learning approaches. The results demonstrate that the proposed model provides a robust data‐driven approach for predicting maximum swelling pressure and that appropriate feature engineering can significantly improve model performance. These findings indicate that deep learning methods offer an effective and practical tool for solving geotechnical engineering problems.