Abstract: Floor water inrush of Coal seam is a complex dynamic phenomenon of hydrogeology and mining, which is controlled by multiple factors and has very complex nonlinear dynamic characteristics. By adopting appropriate risk assessment methods, floor water inrush can be predicted comprehensively and accurately. This paper selects 100 sets of drilling data from Guoerzhuang Coal Mine, and constructed a classification model for the risk of floor water inrush based on confusion matrix from various aspects such as Ordovician water pressure, aquifer thickness, coal seam thickness, coal seam burial depth, thickness of magmatic rock roof, fault scale index, construction points and annihilation points, and structural fractal dimension. Matlab software is used to iteratively train the neural network, determine 64 sets of training samples and 36 sets of testing samples, and establish an optimized MATRIX model. The model divides the risk of water inrush from the floor of Guoerzhuang mining field into four levels: safe, relatively safe, relatively dangerous, and dangerous. Based on the evaluation levels of different coordinates in the mining field, a zoning map of the risk of water inrush from the floor of coal seam 9 was drawn. The corresponding positions and areas of each level are clearly visible in the map. Compared with traditional classification methods based on pressure coefficient and water inrush coefficient, the two criteria are significantly different, but there is no greater than one level in the evaluation results of each region. The model does not have an absolute advantage in any factor, fully leveraging the control ability of each factor. Moreover, multiple indicators such as accuracy, precision, recall, F1 score, P-R curve, and ROC curve of the model are higher than expected values. The prediction accuracy of the partition is high, which solves the problem of data poverty and few factors considered in the “double coefficient” evaluation, enhances the credibility of the partition for water inrush evaluation, and confirms the rationality of the confusion matrix model for water inrush risk classification.