Fraud Detection in Electric Energy Using Differential Evolution

This work proposes the use of deferential evolution algorithm to find the parameters of a data mining system used to pre-select electrical energy consumers with suspect of fraud. A pattern recognition system was built in order to identify suspicious behavior of electrical energy consumers. However, the system only indicates such clients, and the frauds must be confirmed through in-locus inspection. For that reason, it is important that true alarms be high to justify the trade-off of the in locus inspection. Therefore, the parameter of the pattern recognition system must be well tuned, and that can be modeled as an optimization problem using the available training data. This work describes the pattern recognition system in details, and shows the algorithm modeling as an optimization problem. The differential algorithm will be described and results will be show. Results confirm that this approach is feasible.