Effect of multicollinearity in assessing the compaction and strength parameters of lime-treated expansive soil using artificial intelligence techniques

The current investigation introduces the optimal performance models for predicting the optimum moisture content (OMC), maximum dry density (MDD), and confined compressive strength of lime-stabilized soil by comparing support vector machine (SVM), Gaussian process regression (GPR), decision tree (DT), ensemble tree (ET), and multilinear regression (MLR) models. 116, 15, and 14 datasets have trained, tested, and validated these models. For the first time, the liquid limit (LL), plastic limit (PL), fine content (FC), and lime content (LC) have been used as input variables in predicting the compaction and strength parameters of lime-stabilized soil. The performance of models has been computed by root mean square error (RMSE), mean square error (MAE), and correlation coefficient (R) metrics. The performance comparison demonstrates that model ET has predicted MDD with RMSE of 0.393 kN/m³, MAE of 0.201 kN/m³, and R of 0.9824 in the testing phase, close to the ideal values. Also, model ET has predicted UCS with RMSE of 129.630 kPa, MAE of 91.611 kPa, and R of 0.9669 in the testing phase, higher than SVM, GPR, DT, and MLR models. In the case of OMC prediction, model DT has outperformed with RMSE of 4.515%, Mae of 2.633%, and R of 0.9292 in the testing phase. Furthermore, ET and DT models have achieved over 0.9 correlations in predicting MDD, OMC, and UCS of lime-stabilized soil. Based on the score analysis and regression error characteristics (REC) curve, it has been observed that model ET and DT achieved higher scores (= 45) and less REC (= 1.36E−03 in MDD, 2.95E−03 in OMC, and 7.69E−03 in UCS) value in the testing phase. This research presents that moderate multicollinearity highly influences MLR and SVM models in predicting the compaction and strength parameters of the treated soil. The sensitivity analysis reveals that the plastic limit and fine content are the most influencing variables in predicting OMC, MDD, and UCS of lime-stabilized soil.