Support vector machine algorithm is one of the most popular and effective algorithms in the field of statistical machine learning. The research work of support vector machine is mainly divided into two categories: one is single machine support vector machine algorithm for high efficiency; The other is efficient distributed support vector machine algorithm. However, the training efficiency of the existing fast support vector machine is still very slow. How to solve these problems and further improve the efficiency of support vector machine is still a topic worthy of research. In order to identify the non support vectors that do not work on the separation hyperplane, the concept of direction indicator is proposed to quantitatively analyze the position relationship between support vectors and non support vectors; At the same time, in order to prevent the occurrence of misidentified transactions, a non support vector confirmation algorithm based on voting mechanism is proposed; After eliminating non support vectors, the remaining data are trained by support vector machine algorithm to obtain the final model. Compared with the existing stacked support vector machine algorithm in distributed environment, it improves the efficiency of the algorithm and ensures the accuracy of the algorithm. This paper studies how to get the training model quickly in support vector machine.