Abstract: In order to rapidly and accurately determine the risk level of water inrush during the excavation of water-rich tunnels, an improved RF water-rich tunnel risk prediction model based on AHP is proposed.Based on the analysis of 232 tunnel section water inrush accidents, 13 factors covering hydrogeological conditions, design factors and construction factors were selected as evaluation indicators for water inrush risk of water-rich tunnel, and a evaluation index system of water inrush risk of water-rich tunnel was constructed.A dataset of water inrush accidents in water-rich tunnel was established and preprocessed by machine learning methods.Through the application of the dataset and parameter optimization, the weights of each index of RF model were calculated, and then the weights were optimized by AHP, and a RF-AHP model was established.The prediction results of RF model and RF-AHP model were compared and analyzed, and the RF-AHP model was verified by a case study.The results show that the accuracy of RF-AHP model is 98%, which is superior to the RF model.RF-AHP model has good performance in the prediction of water inrush risk in water-rich tunnel, which can provide a new means for the prediction of water inrush risk in water-rich tunnel.