Incremental feature selection based on uncertainty measure for dynamic interval-valued data

• Published in: International journal of machine learning and cybernetics Vol. 15; no. 4; pp. 1453 - 1472
• Main Authors: Shu, Wenhao, Chen, Ting, Cao, Dongtao, Qian, Wenbin
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.04.2024; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Complex Systems; Computational Intelligence; Computing time; Control; Datasets; Decision making; Efficiency; Engineering; Entropy; Feature selection; Fuzzy sets; Information systems; Mechatronics; Methods; Neighborhoods; Original Article; Pattern Recognition; Robotics; Systems Biology; Uncertainty; Interval-valued data; Incremental algorithm; Feature selection; Conditional entropy; Rough sets
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-023-01977-5

Feature selection is an important strategy for knowledge reduction in rough set. Interval-valued data, as an extension of single values, can better express uncertain information from the perspective of uncertainty measure. However, for applications in the real world, feature values in interval-valued data vary with time evolving. For dynamic interval-valued data, it is time-consuming to employ existing approaches to choose the feature subset because they need to recalculate the interval-valued data from scratch when feature values vary. Motived by this, we research the incremental methods of feature selection in dynamic interval-valued data environment, which can select new feature subset according to previous results. At first, the incremental updatings of θ -conditional entropy are proposed, which measures the significance of candidate features along with the change of feature values of a single object and multiple objects, respetively. On this basis, aiming at dynamic interval-valued data, the homologous incremental feature selection algorithms are put forward. Finally, by comparing the results of feature subset selection between incremental algorithm and non-incremental algorithm on public data sets, it can be concluded that the proposed two incremental algorithms are more efficient and effective, especially, as multiple objects change feature values, two incremental algorithms proposed in this paper have achieved satisfactory results in computing time.