Methodology for fraud detection using rough sets

Abstract

This work proposes a methodology based on Rough Sets and KDD for fraud detection made by electrical energy consumers. This methodology does a detailed evaluation of the boundary region between normal and fraudulent costumers, identifying patterns of fraudulent behavior at historical data sets of electricity companies. Using these patterns, classification rules are derived, and they will permit the detection on the database of electricity companies of those clients that present fraudulent feature. When doing inspections with the proposed methodology, the rate of correctness and the quantity of detected frauds are increased, decreasing the losses with electricity fraud on Brazilian electrical energy distribution companies. One of the problems that Brazilian electrical energy distri- bution companies undergo are the commercial losses resulted from consumers electrical frauds. To decrease these losses, the companies realize in loco inspections to detect such frauds. The inspections are made by technicians that go to the con- sumer unit to evaluate equipments and electricity connections. Usually, company experts indicates which consumer unit must undergo the inspection. This decision is based on factors such as: unit with low consumption rate, high fraud incidence, and others. Since there is a very high number of consumer units, it is almost impossible for the expert to evaluate the behavior of each consumer unit and indicate which ones are suspect of fraud. Also, it is not viable to inspect all the consumer units, seeing that the number of fraudulent consumers is small compared to the total number of consumers. The rate of correct fraud identification of the electrical energy distribution companies goes between 5 to 10%. However, it is known that electrical energy distribution companies keep consumer information on theirs databases. This information can be used for the identification of behavior patterns. When finding a pattern that indicates a fraudulent behavior, the expert can recommend that consumers with this pattern must undergo inspection. The discovery process of these behavior patterns when using databases is called KDD (Knowledge Discovery in Databases) (1). This process