Nonparallel Hyperplanes Classifier For Semi-supervised Classification

Machine learning has very important position in artificial intelligence and pattern recognition research. Support vector machine(SVM) is a machine learning method based on statistical learning theory developed in view of the small sample by Vapnik. This method is becoming more and more extensive research and application because it has good generalization ability and is easy to operate for high dimensional data. Traditional supervised classification method could solve various practical problems efficiently, but it needs to have a large number of samples in markers in order to get enough training samples. In a word, the price is high, the efficiency is low. Therefore, some semi-supervised support vector machine(SVM) classification methods are proposed according to the actual needs. The traditional classification method of the supervision problems can solve various practical problems effectively. However, the boundary hyperpanes of the decision function are parallel so that its popularization is restricted. So we put forward nonparallel hyperplanes classifier for semi-supervised classification. It puts labeled data points and unlabeled dates point coding in a connection diagram and each node represents a data point. If there is a lot of similarities between the two data points, with an edge connecting their corresponding node. we should find the right category to make them with existing labeled data and potential graph structure for unlabeled data and minimize the inconsistency to improve the predicting accuracy of the model. At the same time the performance of our proposed method is better than that of Laplacian support vector machine(SVM) and Laplacian twin support vector machine(TSVM) in numerical experiments. Furthermore, we introduce the feature selection technique into our method and improve the accuracy of the model.