Prediction on compression indicators of clay soils using XGBoost with Bayesian optimization

The determination of the compressibility of clay soils is a major concern during the design and construction of geotechnical engineering projects. Directly acquiring precise values of compression indicators from consolidation tests are cumbersome and time-consuming. Based on experimental results from a series of index tests, this study presents a hybrid method that combines the XGBoost model with the Bayesian optimization strategy to show the potential for achieving higher accuracy in predicting the compressibility indicators of clay soils. The results show that the proposed XGBoost model selected by Bayesian optimization can predict compression indicators more accurately and reliably than the artificial neural network (ANN) and support vector machine (SVM) models. In addition to the lowest prediction error, the proposed XGBoost-based method enhances the interpretability by feature importance analysis, which indicates that the void ratio is the most important factor when predicting the compressibility of clay soils. This paper highlights the promising prospect of the XGBoost model with Bayesian optimization for predicting unknown property parameters of clay soils and its capability to benefit the entire life cycle of engineering projects.