Research on the incremental learning SVM algorithm based on the improved generalized KKT condition

Abstract

Abstract In order to adapt to the classification of the large-scale data and the dynamic data, this paper proposes an incremental learning strategy of SVM called GGKKT–ISVM algorithm based on the generalized KKT condition. The algorithm sets the generalized extension factors by the samples distribution density in order to make the useful samples become new support vectors, and it trains a new classifier. Then this algorithm modifies the classifier secondly, and it can not only keep the historical classification information, also can make full use of the new samples’ information, and structure the classifier that has stronger generalization ability. The experimental results show that the algorithm has a good classification effect.