The inverse problem of support vector machines solved by a new fast algorithm

Inverse problem of support vector machine is a powerful technique for generating decision trees with high generalization capability. How to split a given dataset into two clusters such that the margin between the two clusters attains the maximal is the key to the inverse problem. Margin-merging clustering algorithm has been applied to solve this problem with good performance. After the training data are partitioned by margin-merging clustering method, we used a new algorithm to decrease the number of clusters based on the relationship between a global margin and cluster to cluster margin, the enumeration process is simplified greatly. Several experimental results show that the proposed algorithm in this paper decreases the computing time compared with margin-merging clustering enumeration method.