Feature Selection from the Perspective of Knowledge Granulation in Dynamic Set-valued Information System.

Abstract

Set-valued information systems are the generalized models of single-valued information systems. A feature set in such systems may vary dynamically over time when new information arrives, and the feature subset selected by feature selection algorithms need updating for knowledge discovery under a dynamic environment. Knowledge granulation as a feature measure is an effective way to evaluate the discernibility power of the features. But less effort has been made to investigate the feature selection issue from the perspective of knowledge granulation in dynamic conjunctive set-valued information systems. In this paper, we firstly apply the knowledge granulation for measuring features in the conjunctive set-valued information system. With the variation of a feature set in the system, an incremental approach for updating the knowledge granulation is discussed. Correspondingly, an incremental feature selection algorithm is developed when a feature set adds into and deletes from the system simultaneously. The experimental results show that the feasibility and effectiveness of the proposed algorithm in comparison with existing feature selection algorithms.