Machine learning identifies key factors predicting Chengdu clay compressibility

A new machine learning study has identified water content and liquidity index as the dominant factors predicting the compressibility of Chengdu clay, with LightGBM achieving the highest accuracy. The model reached a test R² of 0.8817, significantly outperforming other algorithms. The research used 338 experimental datasets and a hybrid optimization strategy combining Grid Search and Bayesian Optimization. Statistical validation through 30 iterations of Monte Carlo cross-validation and paired t-tests confirmed LightGBM's superior performance, while Random Forest showed better robustness in spatial cross-validation. The study provides a reproducible framework for regional geotechnical machine learning, with SHAP analysis confirming both key factors have monotonically positive effects. Bootstrap confidence intervals offer practical risk references for engineering applications involving Chengdu clay.