Research on an Identification Method for Gas Disaster Risk Based on the Selective Ensemble Classification Model.

Abstract

To improve the accuracy of gas disaster risk identification, a selective ensemble classification model is proposed based on clustering selection and a new degree of combination fitness (CS–NDCF). First, nine base classifiers for gas disasters are constructed on the training data set, including the backpropagation (BP) neural network classifier, naive Bayes (NB) classifier, K-nearest neighbor (KNN) classifier, logistic regression (LR) classifier, decision tree (DT) classifier, support vector machine (SVM) classifier, SVM classifier with cross-validation (SVMCV), random forest (RF) classifier, and gradient boosting DT (GBDT) classifier. Second, the K-means clustering algorithm is used to cluster the base classifiers according to their classification performance. Then, the best performing classifier in each cluster is selected to compose the first selection set. Third, the degree of combination fitness is used to filter the first selection set again to obtain the optimal base classifier result set. Finally, an ensemble classification model is constructed with the optimal base classifier result set. The experimental results on actual mine monitoring data show that compared with the BP, NB, KNN, LR, DT, SVM, SVMCV, RF, and GBDT classifiers, the accuracy of CS–NDCF increases by 7.34, 34.83, 8.28, 12.94, 5.51, 11.72, 6.47, 1.31, and 1.20%, respectively, and CS–NDCF achieves the best forecasting results. Thus, CS–NDCF is an effective method for identifying gas disasters and has a good application value.