SVMs´ Cooperative Learning Strategy Based on MAS to Data Streams Mining

Normal SVM is not suitable for classification problems on large data sets because of high training complexity. To build a distributed learning framework and apply cooperative learning strategy with multiple SVM classifiers are the good inspirations to data stream mining. In this paper, a SVMs´ cooperative learning strategy based on multiple agent system is proposed according to cooperative and distributional traits of MAS. The date streams on Master agent are partitioned into smaller sections which can be assigned to Slave agents, and the data section of each Slave agent are trained and the support vector set trained are combined according their comparability. The implementation of cooperative learning strategy and the final optimal classifier selection are also given as the pseudo codes. At last, the experiment is designed and carried out, and the results confirm the feasibility and validity of the proposed algorithm.