Online streaming feature selection based on Sugeno fuzzy integral
Amin Hashemi, Mohammad-Reza Pajoohan, Mohammad Bagher Dowlatshahi · 2022
Feature selection is a step in which an optimal subset of features is selected for the learning process. This step is applied to high-dimensional data, so that irrelevant and redundant features are removed from the data. Traditional feature selection methods require the entire feature space. At the same time, in many real-world applications such as online social networks, it is impossible to acquire or wait to get the whole feature space. Therefore, online feature selection methods are considered to handle this challenge. In this paper, we present an online feature selection method based on the concept of Sugeno fuzzy integral. According to this method, the streaming features are examined based on several measures, and these measures are combined based on the Sugeno operator. If the aggregated value reaches a pre-defined threshold, the desired feature is selected; otherwise, it is not considered. To prove the performance of the proposed method, comparisons have been made with five online feature selection methods based on two classifiers. We show that the proposed method outperforms competitive methods based on the experimental results.