Support vector machine(SVM) algorithm has much more advantages than other classify algorithm under high-dimensional, small sample and multi-class situation. But at the same time, the parameters optimization has been one of the main factors restricting SVM effect and there is no clear theory to guide it. For original DAG-SVM algorithm’s long time cost and randomness, an improved algorithm has been proposed；For traditional SVM parameter optimization’s large time consuming and unsatisfactory results, an improved artificial fish swarm algorithm has been proposed；Finally,an improved support vector machine algorithm combined 1vsR-DAG-SVM and IAFSA has been proposed. Simulation experiments confirm that the SVM parameter optimization proposed in this paper is feasible and effective.