Self-organizing maps for anomaly detection in fuel consumption. Case study: Illegal fuel storage in Bolivia

Abstract

Currently, Bolivia is a country that suffers problems due to fuel smuggling caused by the subsidy. To address this problem, the government records the fuel supply of each vehicle through a Radio Frequency Identification (RFID) technology as a control action. However, the massive volumes of stored records does not have any intelligent engine to support tasks of detecting anomalies during the monitoring the consumption of each vehicle that could be possible incidents of illegal fuel storage. Thus, the present work proposes an algorithm to identify anomalies behaviors that may be considered fraud cases. We use the unsupervised machine learning technique, Self-Organizational Maps (SOM), to extract patterns of consumption of vehicles and identify anomalies scores based on its own and group history behavior. According to our results, the proposal detects anomalies with 80% certainty.