Feature selection of unreliable data using an improved multi-objective PSO algorithm

• Published in: Neurocomputing (Amsterdam) Vol. 171; pp. 1281 - 1290
• Main Authors: Yong, Zhang, Dun-wei, Gong, Wan-qiu, Zhang
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.01.2016
• Subjects: Classification; Feature selection; Particle swarm optimization; Unreliable data; Unreliable data; Feature selection; Particle swarm optimization; Classification
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2015.07.057

Due to the influence of environment, data obtained in real world are not completely reliable sometimes. This paper focuses on tackling the feature selection problem with unreliable data. First, the problem is formulated as an multi-objective optimization one with two objectives: the reliability and the classification accuracy. Then, an effective multi-objective feature selection algorithm based on bare-bones particle swarm optimization is proposed by incorporating two new operators. One is a reinforced memory strategy, which is designed to overcome the degradation phenomenon of particles. Another is a hybrid mutation, which is designed to improve the search ability of the proposed algorithm. Finally, two state-of-the-art multi-objective optimization algorithms are also applied to this kind of problem, and comparison results suggest that the proposed algorithm is highly competitive for the feature selection problem with unreliable data.