Uncertainty SVM Active Learning Algorithm Based on Convex Hull and Sample Distance

In the process of traditional supervised learning, it is difficult to obtain large numbers of labeled samples and challenging to reduce the cost of data labeling. In response to this issue, and combining the convex-hull vector and the mechanism of support vector machine (SVM), an SVM active learning algorithm based on convex-hull vector and sample distance was proposed. Through the sample distance and convex-hull vector, the algorithm could actively select the samples most valuable to the current SVM classifier, i.e., the samples most likely to be support vectors. The experimental results demonstrated that with no negative impact on the classification accuracy, the proposed algorithm demanded significantly fewer labeled samples compared to random sampling, which reduced the sample labeling cost in learning, enhanced the SVM generalization performance, and increased the training speed.