The classification of tunnel surrounding rock is the basis of tunnel design and construction, which directly affects the safety and operation of the tunnel. In order to realize the rapid and accurate classification of tunnel surrounding rock, a multi-classification softmax regression classification method (MSR) was established based on the softmax regression linear classification model in machine learning. First, the six classification indexes, including rock hardness, rock mass integrity, discontinuity plane occurrence, groundwater condition, joint weathering condition, and initial in-situ stress state, are comprehensively considered and quantified. Second, the argmax function is used as the decision function to establish a multi-classifier. Then, the training samples were corrected by experts and the program is written in Python language to learn the optimal discriminant function. Meanwhile, the accuracy of the model under different learning rates is compared. Finally, the test set data is imported, and the model is automatically calculated to obtain a reasonable classification result of the surrounding rock. Combined with the surrounding rock classification of the Naqiu Tunnel BQ method, the research results show that:① This algorithm is feasible and has high accuracy, which proves that applying machine learning to tunnel engineering can improve the construction efficiency; ② Compared with the binary classifier, multi-classifiers are better suitable for the classification of surrounding rocks;③ When the learning rate is 0.01, the classification performance of the model is the best.