Study on the prediction of loess collapsibility based on feature selection.

【Objective】 For selecting the prediction variables of loess collapsibility from the apparent resistivity,shear wave velocity,coefficient of thermal conductivity,and specific heat capacity of loess,the relationship between loess collapsibility and the indices such as apparent resistivity,wave velocity,coefficient of thermal conductivity,heat capacity were discussed.【Method】 The apparent resistivity,thermal conductivity,and heat capacity of loess samples from a collapsible site were performed.Shear wave velocity test was performed in-siut with the same depth of the loess samples.With feature selection technology,the feature(prediction variables of loess collapsibility) was selected by methods of principal component analysis,CART decision tree and so on.【Result】 The results of feature selection indicated that apparent resistivity,shear wave velocity combined with buried depth could be used as prediction variables.Through prediction of loess collapsibility with prediction variables selected in practical engineering,the result showed relative error of loess collapse settlement was-9.7%.【Conclusion】 The method of predicting loess collapsibility with apparent resistivity,shear wave velocity,and depth of collapsible soil layers is feasible.