A novel method for forecasting time series based on directed visibility graph and improved random walk

Recently network-based method for forecasting time series has become a hot research topic. Although some proposed network-based methods achieve good performance in forecasting some series, how to mine more information of time series and make more accurate predictions is still an open question. To address this issue, we propose a novel reconstructing–forecasting method based on directed visibility graph and random walk process. Firstly, the observed time series is reconstructed to explore more information of series. Then, the reconstructed series is converted into a directed visibility graph. Afterwards, the reconstructed series is predicted with the similarity distribution obtained from improved random walk process. Eventually, the prediction of original time series is calculated using the predictions and the similarity distribution of the reconstructed one. To test the forecasting performance, the proposed method is applied to forecast construction cost index ( CCI ), China’s quarterly total GDP growth ( GDP ) and China’s tertiary industry quarterly GDP growth ( TI ). The results of experiments indicate that, with good robustness, the proposed method is of ability to provide more accurate predictions than compared methods. • An improved visibility graph called directed visibility graph is proposed. More information of time series can be retained in network converting process. • A reconstruction algorithm method is introduced in this paper to approach higher forecasting accuracy. • Paths in random walk process are redefined, which makes it suitable for directed network. • Experimental results on real-world time series demonstrate the effectiveness of the proposed method.