ANOMALY FUEL FRAUD DETECTION TECHNIQUES AND FUEL MONITORING IN LOGISTICS

The logistics and trucking industry continues to face significant operational challenges due to high fuel costs, which typically constitute 20-30% of total operational expenses. Fuel fraud, including fuel draining, side fueling, and unauthorized diversions, further escalates these costs, particularly for Uzbekistan-based IT Park companies managing dispatching and back-office operations for North American heavy-duty trucking fleets. Traditional rule-based monitoring systems often fail to detect sophisticated and subtle fraud patterns hidden within normal operational day. This thesis proposes a hybrid anomaly detection framework that integrates Isolation Forest and Autoencoders for effective real-time fuel fraud detection. The model leverages telematics data, fuel purchase transactions and Transport Management System (TMS) records. Autoencoders are employed to learn complex normal consumption patterns and temporal dependencies in time-series data, while Isolation Forest efficiently isolates anomalous instances in the learned feature space. This hybrid approach enables accurate identification of common fuel theft scenarios, including gradual fuel drainage over multiple stops, side fueling, and unauthorized tank transfers.