Prediction of Shipping Cost on Freight Brokerage Platform Using Machine Learning

Not having an exact cost standard can present a problem for setting the shipping costs on a freight brokerage platform. Transport brokers who use their high market position to charge excessive commissions can also make it difficult to set rates. In addition, due to the absence of a quantified fare policy, fares are undervalued relative to the labor input. Therefore, vehicle owners are working for less pay than their efforts. This study derives the main variables that influence the setting of the shipping costs and presents the recommended shipping cost given by a price prediction model using machine learning methods. The cost prediction model was built using four algorithms: multiple linear regression, deep neural network, XGBoost regression, and LightGBM regression. R-squared was used as the performance evaluation index. In view of the results of this study, LightGBM was chosen as the model with the greatest explanatory power and the fastest processing. Furthermore, the range of the predicted shipping costs was determined considering realistic usage patterns. The confidence interval was used as the method of calculation for the range of the predicted shipping costs, and, for this purpose, the dataset was classified using the K-fold cross-validation method. This paper could be used to set the shipping costs on freight brokerage platforms and to improve utilization rates