Determination of Compaction Parameters of Cement-Lime Soils: Boosting-Based Ensemble Models

Abstract

This study investigates the application of artificial intelligence (AI) models to predict soil compaction characteristics, specifically maximum dry density (M_DD) and optimum moisture content (O_MC), which are critical for stabilizing construction foundations. Traditional methods for determining M_DD and O_MC are labor-intensive and often influenced by factors such as soil type, plasticity, and compaction energy (E). The research employed AI models, including random forest regression (RF-R), gradient boosting regression (GB-R), XGBoosting regressor (XGB-R), and multilinear regression (ML-R), trained on a comprehensive dataset of soil properties. For the first time, compaction energy has been used as an input variable to predict soil cement lime stabilized compaction parameters. Among the models, GB-R demonstrated the highest prediction accuracy for M_DD and O_MC, outperforming RF-R, XGB-R, and ML-R. The performance of built-in models has been measured by three new index performance metrics: the a20-index, the index of scatter (IS), and the index of agreement (IA), in addition to four common metrics. Taylor diagrams confirmed the robustness of these predictions during lab testing. A sensitivity analysis revealed that M_DD and O_MC were most influenced by plastic limit (PL), compaction energy (E), liquid limit (LL), and plasticity index (PI). Additionally, curve-fitting techniques were applied to model the relationship between M_DD, O_MC, and these key factors. The results indicated that the GB-R model, particularly when focused on essential features, provided superior accuracy compared to traditional regression methods, offering a reliable tool for soil stabilization assessments in construction.