An Efficient Network Method for Time Series Forecasting Based on the DC Algorithm and Visibility Relation

Junyin Zhao,Hongming Mo,Yong Deng

doi:10.1109/access.2020.2964067

Junyin Zhao, Hongming Mo + Show 1 more

Open Access

https://doi.org/10.1109/access.2020.2964067

Copy DOI

Abstract

Recently time series prediction based on network analysis has become a hot research topic. However, how to more accurately forecast time series with good efficiency is still an open question. To address this issue, we propose an efficient time series forecasting method based on the DC algorithm and visibility relations on the vertexes set. Firstly, the time series is mapped into the network by the DC algorithm, which is a more efficient approach to generate the visibility graph. Then, we use the variation trends (slope) of those nodes that have visibility relation with the last node to get the preliminary predictive values. Afterward, the value of the last node is adopted to obtain the revised predictive values, which are assigned different weights according to node degree and time distance to get the final weighted result. To better demonstrate the prediction performance and applicability of the proposed method, the proposed method is applied to different time series data sets. The empirical results show that the proposed method could provide a higher level of forecasting accuracy than many methods with relatively lower time complexity.

Highlights

Time series refers to the series of values of the same statistical index in the order of their occurrence
PREDICTION RESULTS AND DISCUSSION To test the predictive performance of the proposed method, we apply the proposed method to predict the values of three different time series, including the Construction Cost Index (CCI), the student enrollment of the University of Alabama, and the price of State Bank of India (SBI)’s share at BSE, India
The decrease of root mean square error (RMSE) and mean absolute percent error (MAPE) shows that the proposed method could improve the accuracy of enrollment prediction in this experiment compared with these listed fuzzy time series based methods

Summary

Introduction

Time series refers to the series of values of the same statistical index in the order of their occurrence. There are many data sets in the form of time series, like temperature [1] and precipitation [2], and most of the economic data are given as time series, such as stock prices [3] and construction costs [4]. One of the primary purposes of time series analysis is to predict the future data based on the historical data, like forecasting financial [5]–[7] and functional time series [8], [9]. How to improve the accuracy of time series prediction has attracted many researchers’ attention. To improve the accuracy of time series prediction, researchers have proposed various time series forecasting methods, such as simple moving average (SMA) [10] and Holt-exponential smoothing (Holt ES) [11]. Combining with autoregressive (AR) [12] and moving average (MA), an autoregressive integrated moving average (ARIMA) [13] forecasting model is built with good

Objectives

Results

Conclusion