Integrating Pattern Growth Mining on SQL-Server RDBMS

Most of the FIMI implementations are memory-like. However when we deal with data warehouse the size of datasets are extremely huge for memory copy. However, using a pure SQL-like approach is inefficient, even almost implementations rarely take advantages of SQL-Extensions such as SQL-Cursors, Store Procedures and UDF functions. Furthermore, RDBMS vendors offer a lot of SQL-Extensions for taking control and management of the data. We propose a pattern growth mining approach tightly coupled on RDBMS for finding all frequent itemsets. The main idea is to avoid one-at-atime record retrieval from the database, saving both the coping and process context switching, the candidate set generation test (expensive joins), and table reconstruction. The experimental results obtained show that our approach is competitive with the most known SQL-like approaches. Our performance evaluation was made with SQL Server 2000 (v.8) and T-SQL.