Governments at all levels are faced with the challenge of setting agendas and creating the frameworks to improve the lives of citizens. As we all know, government touches each of us daily. Policies and initiatives come and go, each seemingly different from the last. Yet, they all have one thing in common: They all cost money, often a great deal of money. The undeniable truth of government is that it is an expensive business; typically, the more radical the agenda or the more holistic the policy, the more it will cost. Another sad truth is that where there is money, there is also fraud, abuse and error. Such misuse—intentional or not—costs the government and, ultimately, the taxpayers.
Consider for a moment fraud in the government. If we accept that government exists to serve the people, to “improve our lives,” then fraud against the public purse removes some of the funding that could improve the lives of all of us, especially those most in need of government support. Not only is government fraud morally, ethically and legally wrong, it is the antithesis of everything good government stands for. Billions of revenue paid by hardworking citizens are lost each year due to improper payments, fraud, waste and abuse. Governments at all levels—federal, state and local—face the enormous challenge of rectifying this situation. What can agencies do to improve collection rates? How do they increase the productivity, effectiveness and efficiency of their auditors and investigators? By identifying a prioritized list of accounts that have a high likelihood of being fraudulent, agencies can optimize investigators’ time and increase the funds collected.
Implementing a strategy and technology solution to find improper payments, fraud, waste and abuse helps governments ensure that vital services and programs that citizens desperately need are there for them. While fraud, waste and abuse have been identified as areas in which data mining is applicable, actually using data mining techniques for this application has historically relied on flagging cases where there are known problems, building models (or profiles) of these problems and “scoring” new data based on the profiles. Although this approach is useful, it is inappropriate to use only this technique because of the following limitations:
1. A situation in which there are two identical records. Problem behavior is sometimes found in one record, while the other record is not even checked. This results in noisy data and problems with model building.
2. The limitations of pattern recognition techniques. Pattern recognition techniques are only able to find patterns that have been found in the past. This means that the existing fraud, waste and abuse often remains undiscovered.
3. Small numbers of identified cases of fraud, waste and abuse. When numbers are low, reliable models cannot be built.
4. Offender behavior changes. Once offenders realize that a certain behavior triggers a problem, they no longer commit that behavior. Then models have to be built to capture the new offending behavior.
Fraud detection is generally hampered by the need for high-skilled investigators plowing their ways through backlogs of computer data, with successive findings triggering new questions involving new painstaking searches; in the meantime fraud, waste and abuse continue. This paper discusses systematic approaches to detecting fraud in three broad categories: vendor fraud, diversion of public funds, and service consumer fraud. Part 2 of fraud detection in government will provide an overview of public sector fraud schemes, then discuss traditional methods and data mining techniques for fraud detection, as well as the implementation of monitoring and reporting systems.