Machine learning regression and classification algorithms utilised for strength prediction of Improved clays

This study applies various machine learning (ML) models to predict soil strength after part-substituting OPC with PFA and GGBS. Stand-alone ML models (BLR, REG, ANN, LR), tree-ensemble ML models (BDT, RDF, DJ), and meta-ensemble ML models (VE, SE) are used. Regression analysis shows higher variance in OPC-substituted predictor variables compared to OPC-alone soils. REG model provides accurate strength predictions (RMSE 0.39, R2 0.86), while tree-based models and meta-ensemble models perform even better (RMSE 0.33-0.06, R2 0.90-0.91). For ML classification, mANN performs well (accuracy 0.78, precision 0.67, recall 0.67) but falls short compared to meta-ensemble models (accuracy 0.80, precision 0.70, recall 0.71). Ensemble methods, particularly VE, demonstrate better performance across different validation methods, providing robust predictions for regression and multiclass classification problems.