Precursory steps to mining HCFA health care claims

Data mining can be used effectively to detect health care fraud and abuse. First, data mining has been successful at applying visualization to very large data sets and recognizing new and unusual patterns of activity. Second, data mining has allowed us to better direct and utilize limited health care fraud detection and investigative resources by recognizing and quantifying the underlying indicators of fraudulent claims, fraudulent providers, and fraudulent beneficiaries. A large amount of work must be performed prior to the actual data mining; probably about 80% percent of our time is spent getting ready to perform the actual data mining. These precursory tasks include: customer discussions, data extraction and cleaning, transformation of the database, and auditing (basic statistics and visualization of the information) of the data. The paper describes the tasks associated as they were performed for HCFA (Health Care Financing Administration) in the USA.