Suspicious Driver Behavior Detection System based on GPT Fine-Tuning over Oil Transportation Tracking Data

This thesis introduces a Suspicious Driver Behavior Detection System tailored for the Fuel Transportation domain using the OpenAI GPT model. By training the model on sensor data received from truck telematics, the system identifies anomalies in driver behavior and routes, thereby flagging potentially suspicious trips. This system operates in parallel with existing fraud detection mechanisms, enhancing their capabilities and aiding in the discovery of novel suspicious patterns. The research methodology includes the preprocessing of the sensor data to adhere to OpenAI's token limitations and to reduce the operational costs of the deployed system. The Fine-Tuning API is employed to train the model using this processed data. Once trained, the model evaluates new real-time data to identify and flag unusual behavior. The primary aim of this research is to improve the security levels in fuel transportation, leveraging advanced AI to detect suspicious activities that might elude existing systems and human oversight.