Abstract: The multi-index models of predicting soil liquefaction level can be divided into two types: models based on classification standard of soil liquefaction level, and models based on instance data. In this research, instance data and data produced by stochastic interpolation based on classification standard were used as training samples. Dimension reduction of the samples was conducted through principal component analysis (PCA), and logistic regression model was adopted to describe the relationship between soil liquefaction level and its influencing factors. Hence the PCA-Logistic models were established for the two model types. Case study proves that the PCA-Logistic models are feasible in the prediction of soil liquefaction level. But the prediction result of the second type (which is based on instance data) of PCA-Logistic model is more in line with the actual situation, and especially has more practical value in the presence of more instance data.

Key words: soil liquefaction level, impact factor, prediction, principal component analysis(PCA), Logistic regression model