The electricity distribution network is responsible for supplying energy to consumers in the National Interconnected System, serving 99% of consumers in Brazil. There are two types of losses in this network: technical losses and non-technical losses or commercial losses. In the case of non-technical losses, the focus of this work, the existence of these results in a higher tariff for all consumers, so that the concessionaire can compensate for such reduction in revenue. Non-technical losses are usually associated with fraud (meter tampering or deviations). The main objective of this work is the application of machine learning techniques, using software R, to identify possible fraudulent behaviors of commercial consumers in the state of Santa Catarina. Considering data from typical consumer load curves and functional information from the company. Preliminary results, using real data from consumers, indicate that the SVM classifier used performed well in the cases studied, achieving precision and accuracy greater than 90%. The input variables selected for the classifier, based mainly on data and information from typical load curves, are the differential of this work, as well as the main reason for the success in the initial tests.