Selecting Samples and Features for SVM Based on Neighborhood Model

摘要

Support vector machine (SVM) is a class of popular learning algorithms for good generalization. However, it is time-consuming in training SVM with a large set of samples. How to improve learning efficiency is one of the most important research tasks. It is known although there are many candidate training samples in learning tasks only the samples near decision boundary have influence on classification hyperplane. Finding these samples and training SVM with them may greatly decrease time and space complexity in training. Based on the observation, we introduce neighborhood based rough set model to search boundary samples. With the model, we divide a sample space into two subsets: positive region and boundary samples. What's more, we also partition the features into several subsets: strongly relevant features, weakly relevant and indispensable features, weakly relevant and superfluous features and irrelevant features. We train SVM with the boundary samples in the relevant and indispensable feature subspaces, therefore simultaneous feature and sample selection is conducted with the proposed model. Some experiments are performed to test the proposed method. The results show that the model can select very few features and samples for training; and the classification performances are kept or improved.