The problem of uncertainty and fuzziness of the mapping relationship between influencing factors of sand liquefaction discrimination and sand state was discussed in this paper. Based on the factor reduction of neighborhood rough set (NRS), a support vector machine (SVM) discriminant model of sand liquefaction was constructed by using the improved sparrow search algorithm (ISSA) with multi-strategy integration to optimize the parameters C and g. 42 groups of examples in Songyuan area, Jilin Province were taken as the overall sample set, including 35 groups as the training set and the other 7 groups as the test set, and nine influencing factors were reduced to four factors by using the NRS. The ISSA-SVM model was then used for prediction, and the sensitivity analysis of the four factors was carried out. The results show that factor reduction can eliminate redundant attributes and reduce the complexity of the model; ISSA algorithm has strong exploration, convergence, and local escape ability. As compared to other models, the NRS-ISSA-SVM discriminant model for sand liquefaction has higher accuracy and stronger gene-ralization ability. To accurately distinguish the liquefaction state of sand, the four factors, i.e., the water level buried depth, the seismic intensity, the standard penetration number, and the thickness of non-liquefied soil, need to be accurately identified, especially the first three factors. The NRS-ISSA-SVM discriminant model for sand liquefaction proposed in this paper can not only accurately judge the sand state of other unknown points in the area, but also provide reference for other similar problems.