Clustering Categorical Data based on Information Loss Minimization

As the size of databases continues to grow, understanding their structure gets more difficult. This, together with the lack of documentation and the unavailability of the original designers of the database adds further difficulty to the job of researchers and professionals to understand the structure of large and complex databases. At the same time, data sources are distributed over several sites and their integration introduces anomalies and often results in “dirty” databases, i.e., databases that contain erroneous or duplicate data records. Our research focuses on the application of data mining, and in particular clustering techniques, to aid the process of recovering and understanding high-level views of data sets.