MRC: Multi Relational Clustering approach

— Clustering is a process of partitioning data objects into groups based on the similarity measures. Most of the existing methods perform clustering within a single table, but most of the real-world databases, however, store information in multiple tables. We propose a new method which is called Multi Relational Clustering (MRC) for clustering a relational database. The MRC approach uses existing clustering algorithms for clustering every table of database. Tables in a database are related to each other based on foreign keys. The MRC approach divides the tables into two categories: dependent and independent tables. A dependent table is a table that includes entities attributes, as well as fields related to the other entities which belong to the other tables. In fact a dependent table includes one or more foreign keys. The MRC approach firstly, clusters independent tables then utilizes these results for clustering dependent tables. The MRC clusters each table by existing clustering algorithm with respect to its fields. An important feature of the MRC approach is ability of clustering several tables in parallel. The proposed approach is very simple and is developed under SQL very efficiently. We offer a version of implementation of k-Means in SQL and use it for clustering a database by MRC approach. Our experiments show that the MRC is efficient for clustering a huge database in a relational environment.