Abstract: Due to the improper foundation treatment of geological disaster prevention and control project in collapsible loess area,the non-uniform collapsibility of loess area poses a certain threat to the prevention and control project of geological disaster. Therefore,selecting appropriate parameters to establish a loess collapsibility prediction model can provide a theoretical basis for the basic design of geological disaster prevention and control projects in loess areas. In this paper,the loess of Gongliu County in the Ili River Valley is taken as the research object. On the basis of collecting a large number of geotechnical test parameters in the area in the early stage,the correlation between the loess collapsibility coefficient and the soil index parameters in the area is analyzed by means of mathematical statistics. The prediction model of loess collapsibility evaluation in the area is established using multiple linear regression theory and neural network theory. The results show that the microstructure of the soil in the study area is the mostly flocculated structure,mainly in the way of support contact. The mineral particles are mostly flaky,and the pore structure is mostly porous or irregular. The material composition is mainly sandstone,albite,calcite and dolomite. The correlation coefficient between loess collapsibility coefficient and these of water content,density,dry density,saturation,void ratio and porosity in the study area is between 0.645 and 0.857,which has strong or extremely strong correlation. Through the comprehensive comparison of the loess collapsibility multiple linear regression model and the RBF neural network model established in the study area,the RBF neural network model is more applicable,credible and accurate,and its accuracy reaches 94.20%. Therefore,the accuracy of the established RBF neural network model can meet the needs of practical engineering,which provides a new idea for solving the problem of collapsibility evaluation of loess in this area.