Fraud investigation at scale: Methods and tools

点击翻译按钮获取中文摘要

Fraud detection is critical for many organizations, such as insurance companies, financial institutions, governments, police, law enforcement units, and retail companies. Typically, after a fraud detection system returns suspects human investigators have to examine those suspects manually. More often than not, investigating a fraud suspect can be very costly. A fraud investigator as a decision maker has to strive for accurate decisions that are effective for business. we propose a focusing research program to tackle the next strategically important challenge in fraud detection: how can we take a systematic data science approach to support human investigators to investigate fraud suspects effectively?This project aims at producing new methods and tools for organizations so that better decisions can be made in fraud investigation to support the business of those organizations. For example, investigating similar suspects together may help to not only lower down the average cost of investigation, but also improve the investigation quality due to more available data and knowledge learned and transferable among similar cases. Investigations can also be planned better so that the gain from the recovered cases can be maximized under various constraints in business, such as within a budget on investigation cost and the amount of extra interaction with customers.This project directly addresses the target area Information and Communications Technologies (ICT), and focuses on the research topic Advanced Data Management and Analytics. Particularly, the project is wholly about Analytics for decision-making. The core theme of the project is to make better decisions in fraud investigation in the context of heterogeneous data at scale, complex business objectives and practical constraints in business. This project closely connects with well engaged collaborators crossing multiple industry segments. It integrates fundamental research, data science tool building, application prototypes and case studies. It also contains a significant HQP training component.