A novel network-based and divergence-based time series forecasting method

Qiuya Gao, Tao Wen, Yong Deng

Abstract: Time series forecasting becomes important due to its wide application in many fields. A variety of methods have been developed to address this problem based on different information in the time series. In this paper, a novel approach is proposed to accurately forecast the time series based on the visibility graph. The similarity between nodes is measured by the topological structure information based on the Jensen-Shannon divergence. The information from top-φ<math><mrow is="true"><mi is="true">φ</mi></mrow></math> most similar nodes is considered to determine the final predicted value with the weighting coefficient obtained by the Gaussian membership function. Two real-world time series data sets are applied to demonstrate the applicability of our proposed model, and results show that our proposed method can achieve lower error indicators compared to other existing network-based approaches.