Abstract: Accurate prediction of railway freight loading and unloading time can improve the scheduling rationality and service quality of railway freight systems, but freight loading and unloading time is affected by various factors. Aiming at the problem of railway freight loading and unloading time prediction, this paper excavated the relationship between freight bill attributes and freight loading and unloading time from the entire process information of freight bills, based on the classification and regression tree model, constructed a gradient boosting decision tree model under the LightGBM framework. The paper integrated, logarithmically transformed, and added features to the relevant data of the entire process of railway freight waybill information to form a waybill dataset, using this dataset to train the constructed model. The results show that the prediction performance of the constructed model for freight loading and unloading time is superior to other machine learning models compared. When this model was applied to actual freight handling business scenarios, the actual accuracy was still higher than other comparison models.