Abstract: The study of loess collapsibility problem has great engineering significance and value. Based on natural moisture content, the five sets of undisturbed loess samples with water content of 12%, 15%, 18% and 20% respectively have been collected from the Jingyang area of Shaanxi Province. The single-line method has been used to test the collapsibility of loess in this study area, the compressive properties and collapsibility coefficient of loess under different normal stresses have been obtained, the influence of different water content and different pressure on collapsibility of loess have been studied in this paper. The results show that the coefficient of collapsibility increases with the increase of pressure at lower moisture content. On the contrary, at higher moisture content, the coefficient of collapsibility first increases firstly, and then decreases with increasing pressure. But, the relationship between the coefficient of collapsibility and the moisture content is relatively complex. At the same pressure, the collapsibility coefficient peaks at a certain moisture content. According to the curve characteristics of collapsibility coefficient, moisture content and pressure, a regression equation has been established for collapsibility coefficient and moisture content of loess under different pressures. Finally, based on the least squares support vector machine, the loess density, moisture content and pressure have been used as predictors, thus, a collapsibility prediction model for loess has been established. The results show that this model established by the support vector machine can meet the engineering requirements for predicting the collapsibility of loess.