Hybrid structures in time series modeling and forecasting: A review

Abstract

Abstract The key factor in selecting appropriate forecasting model is accuracy. Given the deficiencies of single models in processing various patterns and relationships latent in data, hybrid approaches have been known as promising techniques to achieve more accurate results for time series modeling and forecasting. Therefore, a rapid development has been evolved in time series forecasting fields in order to access accurate results. While, numerous review papers have been concentrated on the use of hybrid models and their advantages in improving forecasting accuracy versus individual models in wide variety of areas, no study is concerned to categorize and review papers from the structural point of view in numerous developed studies. The main goal of this paper is to analyze hybrid structures by surveying more than 150 papers employed various hybrid models in time series modeling and forecasting domains. In this paper, the classification of hybrid models is made based on three main combination structures: parallel, series, and parallel–series. Then, reviewed papers are analyzed comprehensively with respect to their specific features of employed hybrid structure. Through reviewed articles, it can be observed that combined methods are viable and accurate approaches for time series forecasting and also the parallel–series hybrid structure can obtain more accurate and promising results than other those hybrid structures. Besides this paper provides the viable research directions for each hybrid structure to help the researchers in time series forecasting area.