Methods for Fraud Detection with Bipartite Graphs

Fraud has always been a huge problem for companies around the world. Today, digital fraud takes the largest part in all fraud cases. Among anti-fraud measures the fraudsters’ connections detection is really important and effective. Graphs help a lot to implement those measures. The present study focuses on graph-based methods for fraud detection and prevention. As specifically, suggests the new method for fraud detection in taxi aggregator using bipartite graph clustering. For this purpose, scientific research about modern anti-fraud measures and graphs usage in such measures is being done. Then a description of a company’s case and a proposition of a solution which is based on bipartite graph clustering is being made. After that, a graph clustering method according to specific company’s case characteristics and proposed solution is being chosen. The result of the study is a fully developed algorithm that finds fraud communities on given data using the chosen clustering algorithm. The algorithm is written on Python using libraries for graphs: NetworkX, igraph and graphistry. The algorithm’s capacity and efficiency was checked on taxi orders data of many Russian cities, big and small. The consequences of this study may be the implementation of such fraud detection method which is narrow-focused for one business model fraud but still efficient and showing good results on specific cases.