Predictors Extraction in Time Series Using Authorities-Hubs Ranking

Nowadays, modern systems are supposed to store and process large time series. As the number of related variables increases, the prediction of time series becomes more and more complicated, and the use of all variables causes some problems for classical prediction models. In this context, the question that may arise is how to select the most relevant set of predictors for a given target time series. In this paper, we present a prediction process for large multivariate time series forecasting, and propose a novel approach for identifying time series predictors, which analyses the dependencies between predictors and a target variable by using the mutual reinforcement principle between Hubs and Authorities of the Hits (Hyperlink-Induced Topic Search) algorithm, which was originally applied to analyze and rank web pages. The results of our experiments are promising, as the proposed algorithm selects predictors that improve the forecasts of many variables compared to methods currently in use, such as the PCA, Kernel PCA, Factor Analysis, and the FCBF (Fast Correlation Based Filter).