Discrete particle swarm optimization approach for cost sensitive attribute reduction

Attribute reduction is a key issue in rough set theory which is widely used to handle uncertain knowledge. However, most existing attribute reduction approaches focus on cost insensitive data. There are relatively few studies on cost sensitive data. Especially, how to evaluate a cost sensitive reduction algorithm is still an issue needing to be studied further. In this paper, we propose four relative evaluation metrics which can be used to compare and evaluate different algorithms for cost sensitive attribute reduction more conveniently. Moreover, we propose a particle swarm optimization method for cost sensitive attribute reduction problem inspired by its powerful search ability. The proposed approach is tested with three typical test cost distributions and compared with an influential algorithm reported recently on both exiting metrics and proposed metrics. Results indicate that the proposed relative evaluation metrics are effective and convenient. Comparing results also show that the proposed algorithm is effective.