Employing Frequent Pattern Mining for Finding Correlations between Tables in Relational Databases

Knowledge of the dependencies and correlations which exist between data units in a relational database is of paramount importance when designing a distributed database since discovering such correlations would be crucial for reaching an optimal data distribution. The knowledge may also be used to improve the performance of a centralized database through a reorganization process. In this paper, we propose a data mining approach for identifying groups of correlated tables in a relational database schema by considering the work profile of the database. We analyze the query history to find sets of tables that are frequently accessed together. The sets of tables that are found overlap in general and we therefore turn each set of tables into a fuzzy set and determine the degree of membership of each table in each of the fuzzy sets. Our approach offers a high degree of flexibility and can be easily customized to produce result with desired amount of detail. The outcome will be highly valuable for guiding a database administrator in producing a better allocation plan. It also guides the database designer in deriving a well optimized fragmentation plan. Our experiment shows the viability and power of our approach. The proposed approach can also be extended to find correlations between smaller data units such as fragments, rows or columns.