Abstract: For the problem that support vector machine requires input vectors are labeled samples which are difficult to obtain in practical application, the paper proposed a prediction method of coal and gas outburst combining semi-supervised learning and support vector machine. It introduced prediction process of support vector machine and selection of input vectors, and improved co-training algorithm of semi-supervised learning: train two different classifiers of SVM and KNN in the same attributes set, and add the samples with the same label in the training set, so as to take full advantage of unlabeled samples to add information continuously, update labeled samples in the training set, and achieve purpose of intensifying training sets. The test result shows that the improved algorithm has higher accuracy than the prediction method using separate support vector machine.