New Insights on Nontechnical Losses Characterization Through Evolutionary-Based Feature Selection
MetadataShow full item record
Although nontechnical losses automatic identification has been massively studied, the problem of selecting the most representative features in order to boost the identification accuracy and to characterize possible illegal consumers has not attracted much attention in this context. In this paper, we focus on this problem by reviewing three evolutionary-based techniques for feature selection, and we also introduce one of them in this context. The results demonstrated that selecting the most representative features can improve a lot of the classification accuracy of possible frauds in datasets composed by industrial and commercial profiles.