The Research On Algorithm Based On Decision Tree And Application In Prediction Of Water Inrush From Coal Floor

In our country, the problem of water-inrush from coal floor has been affecting people’s lives and property safety. Water-inrush from coal floor is the result of comprehensive action of many factors, so, the traditional prediction methods are difficult to solve this problem. In this paper, the problem of coal floor water inrush prediction is summarized as the data mining classification, introducing the decision tree classification algorithm.The decision tree algorithm belongs to the data mining classification techniques. As a simple and effective classification method, it has been widely applied in various fields. But the traditional decision tree algorithms still have certain deficiencies. On the one hand, the classification accuracy still need further improve. On the other hand, the traditional decision tree algorithms can’t handle the unbalanced data sets. In order to solve the existing problems, the improved classification model was verified by UCI data sets, and finally applied to the actual coal floor water inrush prediction. In this paper, the main innovation points are as follows:(1) A kind of mixed split node classification model based on C4.5and CART algorithm is set up. The study of comparison of several common decision tree algorithms shows the differences and the inner link each other, so a kind of hybrid split node classification model based on C4.5and CART algorithm is set up. The result of experiments shows that the classification model based on hybrid split node algorithm, the classification accuracy of improved classification model has a certain degree of increase, but not too much.(2) A decision tree classification model based on the root node information is built. In order to improve the classification precision of the classification of prediction model at a greater degree, considering that the root node selection plays a important role in constructing a decision tree, n (the number of split attribute) decision trees were set up using each attribute as the root node respectively. The performance of the model was verified by UCI data sets, and the result shows that the classification accuracy is better than the single decision tree.(3) A model of decision tree classification algorithm based on the cost sensitive was designed. In the time of big data, there are a large number of non-equilibrium data. The problem of few-class classification difficultly often appears in the classification based on the classical decision tree to non-equilibrium data. In the practical application, the miscalculation of minority class often brings huge costs, reducing the classification error because of the imbalance data, the theory of cost sensitive was introduced. The misclassification cost was imported into the node split, and evaluating the performance of the model with different index. By adjusting the parameters, the accuracy of the minority class greatly improve based on the cost sensitive model in ensuring the classification accuracy of overall and most of the class.(4) The non equilibrium problem in the water-irruption prediction was solved by using the cost sensitive decision tree model. Compared with the classical C4.5and CART algorithm, the cost sensitive decision tree algorithm has a lot of advantages on the true positive rate and negative real rates, G-mean value, F-measure value in the coal seam floor water-irruption prediction. It’s meaningful to the water-irruption control of the actual coal mine.