Social-Spider Optimization-based Support Vector Machines applied for energy theft detection. (January 2016)

• Record Type: Journal Article
• Authors: Pereira, Danillo R.; Pazoti, Mario A.; Pereira, Luís A.M.; Rodrigues, Douglas; Ramos, Caio O.; Souza, André N.; Papa, João P.
• Is Part Of: Computers & electrical engineering. Volume 49(2016)
• Page Start: 25
• Page End: 38
• Publication Date: 2016-01
• Subjects: Nontechnical losses -- Power distribution systems -- Social-Spider Optimization -- Support Vector Machines; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2015.11.001
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: Social-Spider Optimization for model selection in Support Vector Machines. Three distinct scenarios were evaluated. Proposed approach validated in the context of of theft detection in power distribution systems. Graphical abstract: Abstract: The problem of Support Vector Machines (SVM) tuning parameters (i.e., model selection) has been paramount in the last years, mainly because of the high computational burden for SVM training step. In this paper, we address this problem by introducing a recently developed evolutionary-based algorithm called Social-Spider Optimization (SSO), as well as we introduce SSO for feature selection purposes. The model selection task has been handled in three distinct scenarios: (i) feature selection, (ii) tuning parameters and (iii) feature selection+tuning parameters. Such extensive set of experiments against with some state-of-the-art evolutionary optimization techniques (i.e., Particle Swarm Optimization and Novel Global-best Harmony Search) demonstrated SSO is a suitable approach for SVM model selection, since it obtained the top results in 8 out 10 datasets employed in this work (considering all three scenarios). Notice the best scenario seemed to be the combination of both feature selection and SVM tuning parameters. In addition, we validated the proposed approach in the context of theft detection in power distribution systems.