Because there are many factors affecting coal mine water inrush and they have strong correlation, the prediction accuracy of the model will be affected. Due to the heavy workload and high cost of data collection, how to select features scientifically to improve the accuracy of model prediction has become the focus of this study. At first, this study uses stability selection to select 7 factors which are more important in 22 known influence factors, and then builds three typical machine learning classification forecasting models including random forest, neural network, and support vector machine (SVM) to forecast the data before and after feature selection. The experimental results show that the prediction model is very stable after the feature selection and prediction accuracy can reach 100%, and also decrease the cost of the sample data collection.