Forecasting container throughput with big data using a partially combined framework

Abstract

This study proposes a partially-combined forecasting framework for container throughput based on big data composed of structured historical data and unstructured data. Under the proposed framework, the structured data (the original time series) is firstly decomposed into linear component and nonlinear component. Seasonal auto-regression integrated moving average model (SARIMA) is adopted to capture and forecast the linear component, and a combined model, composed of least squares support vector regression (LSSVR) and artificial neural network (GP), is applied to modeling the nonlinear component. Next, unstructured data is analyzed by an expert system. With the synthesized expert judgment, the forecasts of linear and nonlinear components are integrated into a final forecast. For the illustration and verification purpose, an empirical study is conducted with the data of Qingdao Port. The results show that the model under the proposed framework significantly outperforms its competitive rivals.