A Novel Structural Multiple Birth Support Vector Machine for Pattern Recognition

Abstract

Multiple Birth Support Vector Machine (MBSVM) is widely used in various engineering fields due to its fast learning efficiency. However, MBSVM does not consider the prior structure information of samples when constructing each sub-classifier, which often has a strong impact on classification. For the disadvantage, we introduce the prior structure information of samples into MBSVM, and propose a novel MBSVM with structure information in this paper, which is called Structural MBSVM (S-MBSVM). The S-MBSVM inherits the advantage of fast learning speed of MBSVM, and fully utilizes the prior structure information of samples, thus improving the generalization performance. Experimental results show that the algorithm has better classification performance than the classical MBSVM.