Composite machine learning models for forecasting UCS of stabilized lateritic soils

Accurate prediction of unconfined compressive strengths (UCS) of lateral soil is necessity for effective geotechnical and pavement design. To overcome the problem, the current study presents a composite machine learning model to predict the UCS of lateral soil using soil parameters. Different algorithms viz., Bagging, LSBoost, Random forest, ANN and support vector regression are employed and compared to identify the most accurate model. Conventional lab methods such as UCS, OMC, and MDD are time taking, labor necessary, and impractical for large scale applications. Experimental investigations were performed on lateral soils stabilizing with lime, fly ash, gypsum and ceramic slag, which create dataset of compaction and strength parameters. Further, the feature importance analysis shown, MDD and CBR are most significant predictors of UCS. The mixture of lime, fly ash and gypsum yielded the maximum UCS in comparison with other combinations. Among the models, LSBoost demonstrated superior generalization with minimum error (R2 = 0.96, RMSE = 250 KN/m2). The results further indicate, the combined use of lime, fly ash, and gypsum significantly increases the soil strength due to enhanced densification and cementitious bonding. The results suggested suitable ensemble based ML model for reliable UCS predictions, alternative to time consuming laboratory testing and provides a practical decision-support tool, which is potential applications in design and decision-making in field of pavement and geotechnical engineering.