Express Delivery Quantity Forecasting Based on Time Series Decomposition Model

Online shopping, a significant consumption mode, has spurred rapid growth in courier service demand. Thus, accurate courier transport volume prediction is crucial for courier companies to strategically locate warehouses, enhance transportation efficiency, and cut costs. This study proposes a method introducing an express delivery quantity forecasting strategy, grounded in time series decomposition plots. Involving time series graphs, trend charts, seasonal charts, and residual charts, this approach explores the time series' specific characteristics to establish forecasting models. Utilizing time series decomposition plot features, this study initially selects the Vector Autoregressive Moving Average (VARMA) model for forecasting. The model's accuracy is evaluated through Cross-Validation, showing its computed results surpass most machine learning algorithms in Mean Squared Error (MSE) metrics. Subsequently, the study introduces an additional binary variable to indicate delivery feasibility, thereby increasing the forecasting sample size. It then adopts AutoRegressive Integrated Moving Average (ARIMA) models, conducting analyses and improvements for each module. The dataset is split into training and testing sets chronologically. By comparing MSE values, the study ultimately identifies the most suitable parameters.