Two Kinds Of Active Learning Methods

When querying samples, active learning based on SVM did not consider the impact of noise points and the distribution of a large number of unknown regions. To address this issue, the active learning based on semi-supervised FCM and SVM is presented. First, the training samples are clustered by semi-supervised FCM and SVM is trained by cluster central region, and then SVM active learning on this basis. The proposed method makes full use of unlabeled sample, eliminates the impact of noise samples, and avoids the case that separating hyperplane stay in an unreasonable classification for a long-term. In addition, the performance of traditional semi-supervised clustering is not very high for the special data distribution, so a new method that determining the clustering center region is presented. The proposed method can explore the cluster center region of irregular data distribution and make the determination of central regions become more accurate and reasonable.