Semi-supervised Kernel Based Progressive SVM for Coal Mine Gas Safety Data Classification

摘要

It is a new way to perform coal mine safety warning using gas safety data classification. A large amount of samples are easily achieved while the labeling work is time-consuming. Using unlabeled sample to improve the learning performance is the aim of semi-supervised learning. Most existing semi-supervised methods implement either the cluster assumption or the manifold assumption. However, the performance will degrade if the assumption was not appropriate. In this paper, we proposed a method called semi-supervised kernel based progressive SVM which combines both the cluster assumption and the manifold assumption. Specifically, a semi-supervised kernel which reflect manifold information of the samples was constructed by warping the Reproducing Kernel Hilbert Space, and then the semi-supervised kernel was used in SVM which was based on cluster assumption, finally, a progressive learning procedure was used in the proposed method. Several experiments were taken on public datasets and coal mine safety data, the results showed that, compared to the progressive SVM with supervised kernel and the standard SVM with semi-supervised kernel, the proposed method using semi-supervised kernel in progressive SVM had better performance.