The application of synthetic data generation and data-driven modelling in the development of a fraud detection system for fuel bunkering

Abstract

As industry continues to embrace Industry 4.0, many sectors now seek to automate fraud detection to ensure reduced financial exposure. However, the data-driven models which are commonly used in the development of such ‘digital solutions’ rely on ‘supervised’ learning techniques which require high resolution datasets containing labelled instances of the specific fraudulent activity. In reality, applications such as engineering and manufacturing only have limited datasets which contain such information and recreating the physical conditions surrounding the fraudulent activity is often not practical or is illegal. This paper details a collaborative R&D project undertaken for the fuel bunkering industry; whereby data-driven models were designed to detect fraudulent activity during fuel transfer operations. Synthetic data generation was used to build up high resolution datasets based on field data which contained instances of fraud. The results demonstrate successful synthetic data generation and modelling techniques with high predictive accuracies.

Keywords

• Synthetic data
• Data-driven modelling
• Machine learning
• Coriolis
• Bunkering