Decision Support System for the Prediction of Mine Fire Levels in Underground Coal Mining Using Machine Learning Approaches

Abstract

Mine’s fire evaluation is of broad significance for underground engineering structures, as the mine fires are one of the largest fatal accidents among roof falling, gas explosion, accumulation of gases, falling stones, and others. Safety concerns caused by the mine’s fire are a worldwide problem due to the less attention given on safety measures by the management. This study elucidates a new idea to predict mine’s fire levels by employing several machine leaning techniques. A total of 120 patterns of various mine’s fire influencing parameters, i.e., oxygen (O₂) in percent, nitrogen (N₂) in percent, carbon monoxide (CO) in parts per million, and temperature in degree Celsius, are compiled from a Turkish mine, namely, Adularya coal mine. A futuristic dimensionality reduction mechanism t-distributed stochastic neighbor embedding (t-SNE) has been applied to truncate the implication of original data parameters. Moreover, the k-means clustering algorithm was employed to designate the t-SNE dimensionality depletion database. Consequently, the support vector classification (SVC) has been executed to predict various levels of the mine’s fire. The label data was divided into 70:30 ratios for training and testing stages, respectively. The three within-class classification metrics including recall, precision, and f₁-score were employed to determine the performance of the SVC model. The results revealed that the proposed model has higher credibility to predict the various levels of underground mine’s fire.