Abstract: In order to improve the prediction accuracy of coal spontaneous combustion, a model based on KPCA-Fisher discriminant analysis was proposed to predict coal spontaneous combustion, kernel principal component analysis (KPCA) was used tonon-linear feature extraction for characteristic indexes with higher correlation. The extracted principal components were used as the discriminant factor of Fisher discriminant model. The historical data of coal spontaneous combustion in No. 2 Coal Mine of Xuandong was selected, and the model was trained and tested by extracting training set and test set with the ratio of 3∶1 and the forecast results were compared with traditional FDA, SVM, BPNN method. The results showed that KPCA can extract the characteristic indexes of coal spontaneous combustion effectively, and reduce the information redundancy among the indexes. Using Fisher discriminant model based on KPCA to forecast coal spontaneous combustion is not only simple and feasible, but also with high accuracy.