Generic and Scalable Detection of Risky Transactions Using Density Flows: Applications to Financial Networks

Algorithms based on dense subgraphs have been proven to be highly effective in detecting financial risks, but their widespread use has been hampered by well-design density metrics and high-quality solution of densest subgraph problems. Considering that flow is a natural representation of financial transactions, and the transactions in high-risk are often amount-dense, we propose to use density flows for risky transactions detection. However, trivially enumerating paths in a transfer graph is too costly. To tackle this problem, we design a novel combinatorial optimization solution, called MS-FRM. MS-FRM detects densest flow through “k-Hop density grap”, which is constructed by iteratively calculating densest flow with path length 1 to k for each node. During the iteration, a new and generic density metric is used to measure the weights of flows, and pruning is used to delete leaf nodes and reduce computation cost. The generic metric and k-Hop density graph detection make our algorithm suitable for the varieties of risky scenarios. Extensive experimental results on several real and synthetic datasets demonstrate the effectiveness of our approach compared to dense subgraph algorithms. To the best of our knowledge, this is the first work using density flow to detect risky transactions in financial networks.