Predicting the swelling behaviour of expansive soils using decision tree framework

Expansive soils pose significant geotechnical challenges due to their time-dependent swelling behaviour, which can cause severe infrastructure damage. The ability to predict swell potential accurately is vital for effective designs and risk mitigation. This study combines constitutive modelling with machine learning to enhance the predictive capability of swell behaviour in expansive clays. The constitutive swell model introduced by Liu et al. was evaluated by experimental one-dimensional swell tests, demonstrating its ability to accurately predict time-dependent swelling behaviour under varying surcharge pressures (SPs). A data-driven decision tree (DT) framework was created to relate key swelling parameters of Liu et al. model to fundamental soil index properties such as moisture content, SP, and plasticity index to improve predictive accuracy. Based on these measurable index properties, the model successfully captured the final swell potential of clay samples. Independent validation using literature datasets demonstrated high accuracy, with R2 > 0.97 for predicting final swell potential. The proposed DT framework provides geotechnical engineers with a practical tool to assess final swell potential based on fundamental soil index properties, aiding in better infrastructure design and mitigating the risks associated with shrink-swell behaviour.