During the last decades, Anti-Financial Crime (AFC) entities and Financial Institutions have put a constantly increasing effort to reduce financial crime and detect fraudulent activities, that are changing and developing in extremely complex ways. We propose an anomaly detection approach based on network analysis to help AFC officers navigating through the high load of information that is typical of AFC data-driven scenarios. By experimenting on a large financial dataset of more than 80M cross-country wire transfers, we leverage on the properties of complex networks to develop a tool for explainable anomaly detection, that can help in identifying outliers that could be engaged in potentially malicious activities according to financial regulations. We identify a set of network centrality measures that provide useful insights on individual nodes; by keeping track of the evolution over time of the centrality-based node rankings, we are able to highlight sudden and unexpected changes in the roles of individual nodes that deserve further attention by AFC officers. Such changes can hardly be noticed by means of current AFC practices, that sometimes can lack a higher-level, global vision of the system. This approach represents a preliminary step in the automation of AFC and AML processes, serving the purpose of facilitating the work of AFC officers by providing them with a top-down view of the picture emerging from financial data.

翻译：在过去几十年中，反金融犯罪（AFC）机构和金融机构持续加大力度，以应对以极其复杂方式演变与发展的金融犯罪及欺诈活动检测。本文提出一种基于网络分析的异常检测方法，旨在帮助AFC工作人员应对典型的数据驱动场景中海量信息处理挑战。通过在包含超过8000万笔跨国电汇交易的大型金融数据集上进行实验，我们利用复杂网络特性开发了一款具备可解释性的异常检测工具，该工具能够根据金融监管要求识别可能涉及潜在恶意活动的异常节点。我们确定了一组可提供个体节点有用洞察的网络中心性度量指标；通过追踪基于中心性的节点排序随时间演变规律，我们得以突显节点角色中值得AFC工作人员关注的非预期突变。此类变化在当前AFC实践中难以被察觉，因其有时缺乏对系统高层全局视角的把握。该方法作为AFC与反洗钱（AML）流程自动化的初步探索，旨在通过为AFC工作人员提供金融数据图景的顶层视角，促进其工作效率提升。