Improved prediction of clay soil expansion using machine learning algorithms and meta-heuristic dichotomous ensemble classifiers

Soil swelling disasters pose significant geo-hazard risks, necessitating accurate estimation of soil expansion for infrastructure safety. This study introduces an innovative approach using kernelized machines (BLR, BPM, SVM, D-SVM), linear regressors (REG, LR, ANN), and tree-based algorithms (RDF, BDT). Meta-heuristic classifiers (VE, SE) are also utilized for the first time. Various feature combinations influencing soil swelling behavior are explored. Initial results show BLR deviates the most from actual swell-strain, with REG and BLR slightly outperforming ANN. Meta-heuristic learners (VE, SE) exhibit the best performance (highest R2 value of 0.94, RMSE 0.06% for VE). CEC, plasticity index, and moisture content are identified as important features. Kernelized binary classifiers (SVM, D-SVM, BPM) achieve higher accuracy (average accuracy and recall rate of 0.93 and 0.60) compared to ANN, LR, and RDF. Sensitivity-driven diagnostic tests suggest meta-heuristic models perform best with k-fold validation. It is recommended to apply these concepts during preliminary geotechnical or geological site characterization using the best-performing meta-heuristic models.