Abstract: Currently, the control of face straightness in fully mechanized mining faces combines sensor data such as advancing stroke with manual observations. However, an issue has been identified where sensor data and human operation information are not effectively utilized. To address this problem, an intelligent prediction method for face straightness that integrates sensor data and human operation information was proposed. The support advancing cylinder stroke data, support column pressure data and shearer position data were cleaned, and classified according to the normal advancing stroke control distance and the adjusted advancing stroke control distance. A face straightness analysis matrix was constructed, consisting of the normal advancing stroke control distance matrix and the accumulated advancing stroke control distance matrix. Through feature engineering, feature extraction was carried out on the straightness analysis matrix of the working face, and the feature matrix was generated as a sample, with the working condition type corresponding to the manual control distance to serve as sample labels. The experimental results show that the accuracy of the working face straightness prediction model built by random forest algorithm is the highest, which was 91.41%. A machine learning classification algorithm was employed to establish a prediction model for the face straightness of the current mining cycle. This prediction model was applied to the 2312 working faces at the Gaohe coal mine. The results indicated that during the 30-day period and 115 cutting cycles of the face straightness prediction, achieving an accuracy rate of 81.4%.