In order to predict the self-weight collapse coefficient of loess using physical and mechanical indicators rapidly and effectively,this paper takes the loess of the Central Industrial Integration Zone in the Transformation and Comprehensive Reform Demonstration Area of Shanxi Province as an example.The study utilized geographic detector to analyze the relationship between 13 physical and mechanical property indicators in the research area and the self-weight collapsibility coefficient.Through correlation verification,the main influencing indicators were identified,and a machine learning prediction model was established.By comparing the established machine learning models,their accuracy ranges from high to low as follows:extreme gradient boosting model,random forest model,gradient boosting decision tree model,and decision Tree model.The real effectiveness of the extreme gradient boosting model is 83.24%.Therefore,the extreme gradient boosting model established can meet the practical engineering requirements and has certain reference significance for the study of loess collapsibility and related engineering practices.
维普
万方数据