RESEARCH ARTICLE FRBPSO: A Fuzzy Rule Based Binary PSO for Feature Selection Shikha Agarwal 1 • R. Rajesh 2 • Prabhat Ranjan 1 Received: 26 December 2015 / Revised: 26 July 2016 / Accepted: 30 January 2017 Ó The National Academy of Sciences, India 2017 Abstract Particle swarm optimization and fuzzy logic have shown their fruits for many years across the fields of science. Fuzzy logic acts as an intelligent layer to any conventional system. Recently fuzzy logic has been used to improve the performance of particle swarm optimization (PSO). This paper presents a novel fuzzy rule based binary PSO (FRBPSO) for feature selection to get better classifi- cation and a survey on the PSO fuzzy logic hybrid meth- ods. The results on benchmarking high dimensional microarray datasets show the merits of the proposed FRBPSO method. Keywords Particle swarm optimization  Fuzzy logic  Fuzzy rule based PSO  Classification  Feature selection 1 Introduction High dimensionality is a well-known challenge in which numbers of features are very high when compared to the numbers of samples [1]. Dimension reduction is the com- mon approach to deal with challenges of dimensionality. Many statistical and computational methods have been reported in literature [2–7] for dimensionality reduction. These methods can be grouped into two categories; feature selection and feature extraction. Particle swarm optimization (PSO) [8] is one of these computational approaches for feature selection which has shown its merits in many fields of research due to its cognitive/social behavior, exploitation/exploration capa- bility and faster convergence [9]. Basic PSO is a population based optimization algorithm designed for real valued space. Kennedy and Eberhart in 1997 developed Binary PSO (BPSO) [10] for the discrete binary variables. Despite of many advantages, PSO has some drawbacks of getting into local optimum and stagnation. To overcome these problems, many variants of PSO have been proposed by many researchers. Fuzzy PSO is one of the variants of PSO in which fuzzy logic’s strength of uncertainty han- dling is incorporated into PSO to make it more suitable for the optimum result for different applications. This paper presents a survey on the PSO fuzzy logic hybrids for the last one decade, which reveals that in most of the fuzzy PSO variants only parameters of PSO has been optimized using fuzzy logic. Therefore, in this paper a novel fuzzy rule based binary PSO (FRBPSO) has been proposed in which uncertainty in feature selection is han- dled using fuzzy logic. The results on benchmarking dataset show the merits of proposed FRBPSO. 2 Particle Swarm Optimization Particle swarm optimization (PSO) works based on the sharing/learning of information from the past and mimics the searching for food by a flock of birds [8]. In PSO a swarm is made up of some particles (candidate solutions). Each particle (each bird in the flock) represents & Shikha Agarwal shikhaagarwal@cub.ac.in; agarwal.shikha.ibm@gmail.com R. Rajesh kollamrajeshr@ieee.org Prabhat Ranjan prabhatranjan@cub.ac.in 1 Department of Computer Science, Central University of South Bihar, Patna 800014, India 2 Department of Computer Science, Central University of Kerala, Kasaragod, India 123 Proc. Natl. Acad. Sci., India, Sect. A Phys. Sci. DOI 10.1007/s40010-017-0347-8