Abstract: In order to implement the freight management concept of "surplus oriented transportation" and maximize the efficiency of freight management, this paper used multiple machine learning regression algorithms to construct prediction models for each freight transportation cost, and selected the best model based on the performance of different algorithm models on multiple evaluation indicators on the test set, predicted freight charges and external payments by ticket on a daily basis, in order to gain an early understanding of the daily shipping surplus results. The research results indicate that the Extra Trees Regression (ETR) algorithm performs the best in multiple evaluation indicators, and the model constructed using it can implement more accurate prediction of daily freight earnings. As of April 2024, the prediction model has completed predictions for approximately 10.33 million freight bills related to transportation costs, with an overall prediction error rate of less than 1.7%. It fully utilizes the value of data elements and provides data support for freight benefit analysis and business decision-making.