Influence mechanism exploration and machine learning prediction of loess compression deformation coefficient under...

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Material Science

Background:

• Loess subgrade engineering stability relies on accurate compression deformation coefficient prediction.
• Key factors influencing loess compression include compaction, water content, vertical pressure, and molding methods.

Purpose of the Study:

• To investigate the impact of various factors on loess compression characteristics.
• To develop and compare machine learning models for predicting the compression deformation coefficient.
• To analyze feature importance and interactions using SHAP interpretability.

Main Methods:

• Experimental investigation of loess compression under varying conditions.
• Development of XGBoost (XGB), Support Vector Regression (SVR), Backpropagation Neural Network (BP), and SSA-BP models.
• Application of SHAP analysis for feature contribution and coupling effect quantification.

Main Results:

• Vibration compaction, increased compaction, and reduced water content decrease the compression deformation coefficient by enhancing particle interlocking.
• The coefficient significantly increases with rising vertical pressure in compacted loess.
• The SSA-BP model achieved the highest accuracy (RMSE of 0.138%), outperforming SVR, XGB, and standard BP models.

Conclusions:

• The SSA-BP model offers a reliable tool for predicting the compression deformation coefficient in loess subgrade engineering.
• Vertical pressure is the most influential factor, with significant nonlinear interactions observed, especially with water content.
• Findings provide mechanistic insights for improving loess subgrade design and stability assessment.