Assessment of Swelling Characteristics of Fibre-reinforced Expansive Soil using Bayesian Optimized-machine Learning Models

Abstract This study developed machine learning models to predict the swelling pressure (P S ) and swelling potential (SP) of fibre-reinforced expansive soils using a compiled dataset of 187 experimental specimens collected from published literature. The input variables included fibre characteristics (length, aspect ratio, content, tensile strength), soil consistency and compaction parameters (liquid limit, plasticity index, fine content, maximum dry density, optimum moisture content), and swelling characteristics of untreated soil (P S0 and SP 0 ). Tree-based gradient boosting models and artificial neural networks were implemented and evaluated using an 80% training and 20% testing data split. Model performance was assessed using determination coefficient (R²), root mean square error (RMSE), mean absolute error, and additional statistical indices. The gradient boosting model achieved the highest predictive accuracy, with testing R² values of 0.974 and 0.972 and RMSE values of 14.2 kPa and 0.82% for P S and SP, respectively. Feature importance analysis showed that untreated soil swelling parameters (P S0 and SP 0 ) were the dominant predictors governing swelling behaviour after fibre reinforcement. The results demonstrated that machine learning models, particularly gradient boosting methods, provided reliable and accurate prediction of swelling characteristics, enabling efficient evaluation and design of fibre-reinforced expansive soils.