Research on Logistics Sorting Volume Prediction Based on Machine Learning

In modern logistics, accurate forecasting of sorting center cargo volumes is crucial for resource allocation and process optimization. Traditional time series forecasting models, such as ARIMA and SARIMA, perform well in short-term predictions but encounter diminishing performance in long-term scenarios. This study aims to overcome these limitations by employing non-traditional time series forecasting techniques, specifically machine learning algorithms. We constructed a predictive model that includes a variety of features. By developing a comprehensive set of features that includes lagged cargo volumes, time series characteristics, and trend indicators, we trained individual predictive models for each sorting center. During the model selection process, we evaluated multiple machine learning models, including random forests, neural networks, support vector machines, and linear regression. The evaluation was based on the models' fit to the training set and their performance on the validation set. By analyzing residual plots and calculating scores on the test set, we determined the models' predictive accuracy and bias to select a reliable and precise forecasting model. The model evaluation utilized MAE, MSE, and RMSE metrics, with the random forest outperforming the other algorithms. Ultimately, we used the random forest model to provide a 30-day cargo volume forecast for sorting centers, enhancing predictive accuracy with hourly data granularity.