Experimental study and machine learning prediction on compressive strength of industrial waste- solidified marine soft soil under dry-wet cycles

Abstract

The rapid development of infrastructure in coastal regions has led to the deterioration of soil mechanical properties due to dry-wet (D-W) cycles, which significantly affects the durability of engineering projects. While most previous studies have focused on the impact of curing agents on the mechanical properties of marine soft soil (MSS), a systematic framework for predicting the strength of stabilized materials under D-W cycles is still lacking. To address this gap, this study utilizes blast furnace slag (GGBS), fly ash (FA), and lime in combination to improve MSS. A database containing 624 Unconfined compressive strength (UCS) data points was established to study the strength characteristics and curing mechanism of solidified Marine silt (LGF-MSS) under D-W cycles, and the optimal content of curing agent was determined. Using the XGBoost machine learning framework, optimization algorithms including the Whale Optimization Algorithm, Particle Swarm Optimization, Sparrow Search Algorithm, Grey Wolf Optimization, and Firefly Optimization Algorithm were applied to develop a UCS prediction model under D-W conditions. The SSA-XGBoost model achieves optimal performance in UCS prediction, with a coefficient of determination (R²) of 0.9786 on the test set. In addition, the study provides the importance of curing age, LGF content, number of cycles, degree of compaction, and drying temperature by using correlation analysis, sensitivity analysis, and SHapley Additive exPlanations (SHAP). The developed high-precision prediction model effectively predicts the strength of LGF-MSS under D-W cycles, offering strong technical support and decision-making references for related engineering practices.