Combined Identification Model of Water Inrush Source in Coal Mine Aquifers Based on MIC-FA-Logistic Regression Method: A Case Study of Yangliu Coal Mine, China

Rapid and accurate identification of the source of mine water inrush is a key technical challenge in preventing and controlling such accidents. To enhance the accuracy and stability of source identification, this study proposes a combined model integrating maximum information coefficient (MIC) feature selection, factor analysis (FA) for dimensionality reduction, and Multinomial Logistic Regression. First, MIC is utilized to select key variables from hydrochemical indicators that exhibit strong correlations with the water source type, effectively capturing significant nonlinear characteristics. Second, FA is adopted to reduce the dimensionality of the selected features, eliminate multicollinearity, and extract potential common factors. Finally, a combined discrimination model based on MIC-FA-Logistic regression is constructed. Using the Yangliu Coal Mine in the Huaibei Mining Area as a case study, 67 water samples were collected from four aquifers. Nine hydrochemical indicators, specifically Na+ + K+, Ca2+, Mg2+, Cl−, HCO3−, SO42−, total hardness, pH value, and total alkalinity, were selected as initial variables. MIC effectively quantified the complex correlation strengths between these indicators and the water source types. Consequently, eight indicators (Mg2+, SO42−, Ca2+, total hardness, pH, HCO3−, total alkalinity, and Na+ + K+) were selected as key discriminant variables. FA transformed these eight indicators into five new comprehensive variables, optimizing the model’s input structure. The discrimination accuracy rates of the MIC-FA-Logistic regression model for the training and test samples were 89.1% and 95.2%, respectively. This performance is significantly superior to traditional Logistic regression, FA-Logistic regression, MIC-Logistic regression, and SVM models. This study provides a method for discriminating mine water inrush sources characterized by high precision, high stability, and strong interpretability.