The hybrid PROPHET-SVR approach for forecasting product time series demand with seasonality. (November 2021)

• Record Type: Journal Article
• Authors: Guo, Liang; Fang, Weiguo; Zhao, Qiuhong; Wang, Xu
• Is Part Of: Computers & industrial engineering. Volume 161(2021)
• Publication Date: 2021-11
• Subjects: Product Demand prediction -- Time series demand with seasonality -- SVR -- PROPHET -- Hybrid PROPHET-SVR approach; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107598
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A hybrid seasonal forecasting model based on PROPHET-SVR is proposed. The proposed model performs strongly in capturing seasonal and nonlinear patterns in time series data. SVR residual correction and improved parameter determination method are used to improve the forecasting accuracy. The proposed model outperforms other models under comparison in terms of forecasting accuracy. Abstract: Demand forecasting is the basic aspect of supply chain management. It has important impacts on planning, capacity and inventory control decisions. Seasonality is a common characteristic of most time series demands in practice. Thus, regarding seasons and holidays as important factors of demand forecasting is nontrivial, which contributes to increased forecasting accuracy. In this study, we propose a hybrid approach that integrates Prophet and SVR (support vector regression) models to forecast time series demand in the manufacturing industry with seasonality. In the proposed hybrid PROPHET-SVR approach, Prophet is used to forecast the seasonal fluctuations and determine the input variables of SVR, and SVR is used to capture nonlinear patterns. Therefore, the approach can not only customize the influence of holidays and seasons but also account for the forecasting residual to increase the accuracy. Computational results demonstrate that the hybrid PROPHET-SVR approach outperforms a variety of other prediction methods. This paper also illustrates the application of the new forecasting method in a case of the manufacturing industry in China, and proves the robustness of the method.