A systematic forecasting framework for the key factors of supply chain transportation

Supply chain transportation is an important part of supply chain management. Thus, good transportation conditions can stabilise the production plan of an enterprise, reduce off-production risks, improve operational efficiency, and more. The forecasting of supply chain transportation is a popular research direction with a long history. Many scholars have studied forecasting models to predict key factors of supply chain transportation from various perspectives. In the present study, the forecasting models for key factors of supply chain transportation were systematically explored. Through a series of experiments, a systematic forecasting framework for the key factors of supply chain transportation has been proposed. By accurately forecasting the future value of key factors of supply chain transportation, relevant enterprises can benefit from this framework by adjusting corresponding operating strategies, planning reasonable personnel structure, purchasing a sufficient amount of equipment, reducing labour waste, improving operating efficiency, and so forth. First, the present study involved conducting comparative experiments to examine the performance of different forecasting methods for the same time series of supply chain transportation. In terms of forecasting accuracy, the results indicated that machine learning methods (non-linear methods) are a better choice than simple methods (linear methods), which is in line with previous findings by other scholars. Particularly when the number of observations is limited, support vector regression represents a promising forecasting method. An important message can be obtained from the results is that while machine learning approaches are useful for training forecasting models, the nature of the problems can affect the performance of these approaches. Therefore, they may not necessarily be better than traditional regression-based forecasting approaches. In other words, machine learning approaches should be applied with caution. Second, the present study used numerous data sources to train the forecasting model. Specifically, data on the international business environment was collected. The results indicated that data from other ports in contact with the studied port are also important and can improve forecasting accuracy. This finding highlights a new data source for forecasting research. Third, this study overcame the collinearity issue of the input variables for the forecasting model by employing ridge regression and a support vector machine. The experimental results showed that ridge regression can solve collinear issues in the economic data of the studied port city and select suitable input variables for learning methods. The results also indicated that the clustering of data is a good method for data processing when there are many data types. Fourth, the present study built a hybrid forecasting model for univariate time series forecasting. The seasonal autoregressive integrated moving average (SARIMA) was used to obtain the Gaussian white noise from the raw time series. Then, the support vector machine was used to capture the patterns from the raw time series and Gaussian white noise. Thus, the experimental results suggest that Gaussian White Noise can increase forecasting accuracy by providing more information. The results also indicate that the forecasting results generated by SARIMA cannot be reused to train other forecasting methods