Research On Cargo Volume Prediction at Sorting Centers Based on Multiple Machine Learning Models

Abstract

Forecasting cargo volume is a crucial part of the workflow in sorting centers and has a significant impact on their operational capabilities. This paper considers a variety of machine learning methods to predict future cargo volumes at sorting centers. By analyzing the variance and mean differences in cargo volumes across different centers in the dataset, multiple features were extracted, including historical cargo data, time series features, and individual trend indicators for different centers. These features were used to train the feature sets for Random Forest models, Neural Network models, and Linear Regression models. Additionally, this paper evaluated the prediction performance of these machine learning models using the MAE, MSE, and RMSE metrics. Ultimately, the Random Forest model, which showed the best evaluation results, was selected to forecast future cargo volumes at sorting centers. The predicted results from the Random Forest model showed that for sorting center SC1, the MAE was 0.085, MSE was 0.313, and RMSE was 0.177; for sorting center SC35, the MAE was 0.072, MSE was 0.019, and RMSE was 0.176. These results indicate that the Random Forest model developed in this study is highly accurate in predicting cargo volumes at sorting centers.