Long-Term Trend Analysis and Prediction of Freight Volume in Beijing Using ARIMA-LSTM Model

Abstract

To study the relationship between freight volume in Beijing and various economic factors and reveal the underlying reasons for changes in freight volume, we selected five sets of data highly correlated with freight volume, namely milk production, vegetable and edible fungus production, furniture production, steel production, and cement production in Beijing, as input features for the prediction model based on the results of Pearson correlation coefficient (PCC) analysis. After data preprocessing and feature engineering, we trained these data using the Autoregressive Integrated Moving Average - Long Short-Term Memory Hybrid (ARIMA-LSTM) model and predicted the freight volume in Beijing for the first quarter of 2024. Meanwhile, we established a BP neural network model as a comparison benchmark. By comparing the prediction performance of different models, the results indicated that the ARIMA-LSTM model performed better in handling long-term sequence data and capturing nonlinear relationships, providing more accurate and reliable freight volume prediction results. Finally, we observed a significant downward trend in the freight volume in Beijing in recent years.