Feature Importance Analysis for Compressive Bearing Capacity of HSCM Piles Based on GA-BPNN

To address the complex pile–soil interaction mechanisms in predicting the compressive bearing capacity of HSCM piles (Helix Stiffened Cement Mixing piles) in marine soft soil regions, this study proposes an intelligent prediction method based on a GA-BPNN (Genetic Algorithm-Optimized Back Propagation Neural Network). A high-quality database comprising 1243 data points was established through finite element numerical simulations. By integrating data preprocessing techniques and the GA-BPNN model, the study systematically investigated the influence of helical blade spacing H1 and H2, strength ratio Cref/Su, and diameter ratio Dsc/DH on bearing capacity. The results demonstrate that the GA-BPNN model achieves a prediction accuracy of 99.07%, with a mean squared error (MSE) of 7.20 × 10−3 and a coefficient of determination R2 of 0.990. SHAP value analysis reveals that the strength ratio and diameter ratio are the dominant factors, exhibiting nonlinear relationships with bearing capacity characterized by saturation effects and threshold-dependent behavior. Laboratory tests further confirm strong correlations between cement–soil strength Cref, formed pile diameter Dsc, and bearing capacity. The findings indicate that the GA-BPNN model provides an efficient and accurate approach for predicting the bearing capacity of HSCM piles, offering a reliable basis for engineering parameter optimization.