Advanced Machine Learning Techniques for Predicting Compaction and Strength Characteristics of Cement Kiln Dust-Stabilized Soils

چکیده مقاله

This study used advanced prediction methods to evaluate the unconfined compressive strength (UCS) and compaction parameters, including optimum moisture content (OMC) and maximum dry density (MDD), of CKD-stabilized soil. After preparing a comprehensive database of studies on CKD-stabilized soils, ten machine learning models were developed for modelling, and their performance was compared. Subsequently, further analyses were performed to understand the relationship between independent variables and MDD, OMC, and UCS of CKD-stabilized soil. The results indicated that Categorical Boosting, Random Forest, and Extreme Gradient Boosting were the best-performing models on MDD, OMC, and UCS, respectively. The MDD and OMC of untreated soil and calcium oxide in CKD were determined to be effective parameters for predicting the MDD of stabilized soil, while the MDD and OMC, along with the liquid limit of untreated soil, were identified as important parameters for predicting the OMC of stabilized soil. The UCS of untreated soil, CKD content, and curing time were the most important parameters for predicting the UCS of stabilized soil. Finally, the non-linear relationship between the most important parameters and MDD, OMC, and UCS of CKDstabilized soil was presented. This research provided valuable insights for on-site decision-making and for improving infrastructure development projects.