非均衡样本下基于GRA-BSMOTE-RF的瓦斯突出预测

Gas Prominence Prediction Based on GRA-BSMOTE-RF under Unbalanced Samples

为解决煤与瓦斯突出预测模型由于样本数据不均衡导致的分类效果不好的问题,提出一种将过采样方法(BSMOTE)和随机森林(RF)模型相耦合的预测模型。首先通过灰色关联分析(GRA)进行特征选择。其次,通过BSMOTE方法增加突出的少数类样本,有效地区分类别区域边界。最后,构建GRA-BSMOTE-RF煤与瓦斯突出预测模型,以此来减少类别不平衡对模型预测的影响。根据结果表明,提出的模型对于少数类的分类正确率明显提升,证实GRA-BSMOTE-RF模型在不平衡数据下的煤与瓦斯突出预测上具有较好的预测效果。

To mitigate the issue of suboptimal classification accuracy in coal and gas outburst prediction models stemming from imbalanced sample data,proposes a prediction model that integrates the oversampling method Borderline-SMOTE(BSMOTE)with the random forest(RF)model.Initially,feature selection is conducted using grey relational analysis(GRA).Subsequently,the BSMOTE method is employed to augment the minority class samples of outbursts,thereby effectively distinguishing boundary regions.Ultimately,the GRA-BSMOTE-RF model for coal and gas outburst prediction is established,which aims to alleviate the impact of class imbalance on model prediction.The findings suggest that the proposed model substantially enhances the classification accuracy of the minority class,and demonstrate that the GRA-BSMOTE-RF model performs well in predicting coal and gas outbursts under imbalanced data.