Abstract: Debris flow is a natural disaster in mountainous areas with great damage, whose hazard assessment has a significant meaning in disaster prevention and reduction. Combining with the applications of support vector machine method in debris flow hazard assessment in the past ten years, this paper focuses on the selection of different evaluation indexes for debris flow, the determination of parameters that influences performance for support vector machine, the influence of unbalanced debris flow data on evaluation results and the generalization of debris flow evaluation model. First of all, introducing the theory of rough set for attribute reduction of debris flow evaluation indexes, then filtering the core indexes which influence the evaluation results of debris flow; secondly, comparing the influences of evaluation results caused by penalty factors and kernel function parameters of support vector machine determined by grid search algorithm, genetic algorithm and particle swarm algorithm respectively; finally, by training and testing the measured data of different evaluation indexes and risk grades for single debris flow ditches, establishing the improved support vector machine model of debris flow hazard assessment, making research on the influence of hazard evaluation results caused by the disequilibrium of debris flow data, and applying the established model in different regions for debris flow hazard assessment to make generalization verification. The results show that the SVM model can be used in the debris flow hazard assessment, but the accuracy of evaluation and the strength of generalization ability are closely related to the selection of evaluation index, the determination of performance parameters for SVM and the equilibrium of debris flow data, only the research of related issues for support vector machine model are strengthened in practical application, can we establish the model of debris flow hazard assessment with better applicability.