In response to the difficulty in predicting the change of pre-tightening force when using torque method to load bolt, this paper proposes a bolt pre-tightening force prediction method based on mechanism and data fusion calculation for hexagonal end face bolt, and establishes a tightening prediction model based on machine learning method. Firstly, a tightening mechanism model of any structural bolt is established, revealing the reasons why pre-tightening force is difficult to predict and errors cannot be eliminated. Secondly, sensitivity evaluation indicator is established to conduct parameter sensitivity analysis on the mechanism model, and the fusion method of “mechanism model guiding data model to perform feature selection” is determined. Finally, a bolt tightening prediction model based on Gaussian Process Regression is proposed, and corresponding engineering prediction software is established. The experimental results show that this prediction model can not only predict the variation of pre-tightening force with torque, but also synchronously display the confidence interval of pre-tightening force fluctuation in a probabilistic sense. Under different external conditions, the prediction results can still maintain good consistency with the experimental results. The prediction model breaks through the limitation of traditional method, which calculates the torque coefficient and indirectly loads the pre-tightening force. It is of great significance to improve the accuracy of pre-tightening force and determine the error size.