What is the Importance of Selecting Features for Non-Technical Losses Identif cation?

Abstract

Although non-technical losses automatic identif cation has been massively studied, the problem of selecting the most representative features in order to boost the identif cation accuracy has not attracted much attention in this context. In this paper, we focus on this problem applying a novel feature selection algorithm based on Particle Swarm Optimization and Optimum-Path Forest. The results demonstrated that this method can improve the classif cation accuracy of possible frauds up to 49% in some datasets composed by industrial and commercial prof les.