Abstract: The innovative implementation of horizontal directional drilling technology in ultra-long distance tunnel geological surveys can avoid the constraints inherent in traditional vertical survey methodologies, fundamentally transforming the current status of geological surveying in tunneling projects.This study constructed a prediction dataset of surrounding rock from various drilling parameters collected in the field.After data imbalance process, mud pressure, drilling velocity and bottom-hole drilling pressure were taken as surrounding rock classification evaluation indicators, and the classification performance of each model was evaluated and validated by introducing various machine learning algorithms and grid search for tuning parameters.The classification performance of each model was evaluated and investigated by accuracy, precision, recall, and F1 score.The results showed that:(1) Most of the algorithms had good classification performance for carbonaceous slate and gneissic granite, while the prediction accuracy for quartz gneiss was low.(2) Random Forest, Decision Tree and Gradient Boosting Decision Tree had the best classification performance, while the AdaBoost classifier had the worst performance.(3) Compared with traditional methods, machine learning classification had more advantages in dealing with nonlinear problems.Establishing data-physical dual-driven intelligent classification theory and model for surrounding rocks was one of the development trends.The results of this study can provide substantial insights and guidance for the prognostication and categorization of surrounding rocks in boreholes executed via horizontal directional drilling.