The rise of electronic banking, digital transactions, and finally virtual currencies such as Bitcoin is an indicator for the ongoing digital transformation in financial technologies. Currency units now generated by decentralized systems and can be transferred globally within minutes and with minimal transaction costs. In contrast to existing currency systems, virtual currencies are operating without central control (e.g., national banks) and bypass established payment processors (e.g., banks). All transactions, which have ever been executed with Bitcoin, are anonymous and accessible in the publicly visible blockchain and can therefore be accessed for analytics tasks.
GraphSense aims at developing algorithmic solutions for real-time analytics of digital currency transactions to provide insight into cash flows and functionality of currency systems. A special focus lies on anomaly detection, which should identify transactions and transaction patterns that deviate from typical structures. This could, for instance, help in identifying and tracing fraudulent activities.
Real-time analytics of large network structures
The specific characteristic and scientific challenge of the GraphSense project lies in the structure and volume of the transaction data to be analyzed.
More than 100M atomic transactions form a network in which transactions are represented by hundreds of millions nodes and edges. Anomaly detection algorithms operating on such structures must be built for horizontally scalable infrastructures (e.g., Apache Spark) and tested for their applicability.
Technologies developed within the GraphSense project should, besides the use case “Anomaly detection in virtual and non-virtual digital currencies”, also be applicable for other application areas (e.g., fault detection in manufacturing processes, anomaly detection in energy networks). All developed components will therefore be published as open source software.