Abstract: In order to improve the prediction accuracy of coal spontaneous combustion in caving zone under imbalanced data, after taking O₂ concentration etc. as factors, and the principal components of factors were obtained by PCA, a PCA-AdaBoost prediction model of coal spontaneous combustion was built, which took the principal components as inputs and combustion situations as outputs. Taking Xuandong 2^nd coal mine as the research object, the model was trained through twenty groups of training samples, and evaluated by the area under the curve of receiver operating characteristic curve. The trained model was used to predict fifty groups of test samples. A prediction result comparison was made between the model and the PSO-SVM model. The results show that based on imbalanced data sets, three principal componets are extracted with the 86.831% information of six original factors by PCA, both the correlations between the factors and the dimensionality have been reduced, that temperature and CH₄ concentration have a greater impact than other factors, that the prediction results of PCA-AdaBoost model accord with the actual situation, and that the model is superior to the PSO-SVM model in terms of prediction accuracy and convergence speed.

Key words: spontaneous combustion, imbalanced data set, principal component analysis(PCA), adaptive boosting algorithm(AdaBoost), particle swarm optimization-support vector machine(PSO-SVM)