Weighted bee colony algorithm for discrete optimization problems with application to feature selection. (September 2015)

• Record Type: Journal Article
• Authors: Moayedikia, Alireza; Jensen, Richard; Wiil, Uffe Kock; Forsati, Rana
• Is Part Of: Engineering applications of artificial intelligence. Volume 44(2015:Aug.)
• Page Start: 153
• Page End: 167
• Publication Date: 2015-09
• Subjects: Bee colony optimization -- Categorical optimization -- Classification -- Feature selection -- Weighted bee colony optimization; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2015.06.003
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: The conventional bee colony optimization (BCO) algorithm, one of the recent swarm intelligence (SI) methods, is good at exploration whilst being weak at exploitation. In order to improve the exploitation power of BCO, in this paper we introduce a novel algorithm, dubbed as weighted BCO ( w BCO), that allows the bees to search in the solution space deliberately while considering policies to share the attained information about the food sources heuristically. For this purpose, w BCO considers global and local weights for each food source, where the former is the rate of popularity of a given food source in the swarm and the latter is the relevancy of a food source to a category label. To preserve diversity in the population, we embedded new policies in the recruiter selection stage to ensure that uncommitted bees follow the most similar committed ones. Thus, the local food source weighting and recruiter selection strategies make the algorithm suitable for discrete optimization problems. To demonstrate the utility of w BCO, the feature selection (FS) problem is modeled as a discrete optimization task, and has been tackled by the proposed algorithm. The performance of w BCO and its effectiveness in dealing with feature selection problem are empirically evaluated on several standard benchmark optimization functions and datasets and compared to the state-of-the-art methods, exhibiting the superiority of w BCO over the competitor approaches.