An election strategy for online streaming feature selection

Abstract

Feature selection (FS) is one of the most effective methods in data preprocessing. In many real-world applications, such as social networks, getting all the features or even waiting for them is impossible. Hence, common feature selection methods are not applicable to such data. Thus, online streaming feature selection methods are provided to deal with such data where the entire feature space is not available from the beginning. On the other hand, ensemble methods have recently shown that they can effectively improve the performance of feature selection methods. In this paper, a new method is proposed based on the ensemble of multiple filter rankers to enhance the performance of feature selection methods in an online streaming space. This ensemble process is modeled as an election process, and the Weighted Borda Count (WBC) method is utilized to aggregate the votes. The proposed method showed better classification performance than the experiments' other methods.