Correlation analysis for decision support with applications to law enforcement

Correlating elements of large databases that are related but not exact matches has importance in a variety of applications. In health care epidemiologists have an interest in searching records for patterns of disease. In law enforcement this correlation task enables crime analysts to associate incidents possibly resulting from the same individual or group of individuals. In practice, most analysts perform this task manually by searching through records looking for similarities. Manual search does not imply paper records, but rather the construction of search criteria to narrow the search but not overly restrict it, either. The paper describes automated approaches to record or report correlation. Each of the automated approaches employs a weighted sum of attributes as the total similarity measure (TSM) between any elements in the database. All the approaches build the TSM using prior information provided by experienced analysts. We compare the methods using real data from a law enforcement example.