Incremental approaches for feature selection from dynamic data with the variation of multiple objects

Abstract Owing to the dynamic characteristics of data in the big data era, multiple objects of a decision system often vary with time when new information arrives in real-world applications. However, many feature selection algorithms are designed for static decision systems, and some dynamic feature selection algorithms treat the variation of multiple objects as the cumulative variation of a single object. In an environment where multiple objects vary with time, these algorithms are often time-consuming. Therefore, strategic behaviors need to be reinforced to improve the efficiency of feature selection. Incremental updating is an efficient technique, which can be applied to deal with dynamic learning tasks because it can make use of previous knowledge to obtain new knowledge. In this paper, we focus on the incremental updating to select a new feature subset with the variation of multiple objects. First, the dependency function is updated in an incremental manner to evaluate the quality of candidate features. Then two incremental feature selection algorithms are developed when multiple objects are added to or deleted from a decision system. Experiments on different UCI data sets show that the proposed algorithms can select new feature subset in much less computational time and do not lose the classification performance when compared with other algorithms.