Parallel SQL Based Frequent Pattern Mining

Data mining auf gorßen relationalen Datenbanken hat an popularität gewonnen und seine Besonderheiten sind wohl bekannt.Die Perfomance von SQL gestütztes Data Mining fällt bekannterweise hinter spezialisierten Implementationen zurück.

Data mining on large relational databases has gained popularity   and its significance is well recognized. However, the performance   of SQL based data mining is known to fall behind specialized   implementation. We investigate approaches based on SQL for   the problem of finding frequent patterns from a transaction   table, including an algorithm that we recently proposed, called   Ppropad (Parallel PROjection PAttern Discovery). Ppropad   fundamentally differs from an Apriori-like candidate set   generation-and-test approach. This approach successively   projects the transaction table into frequent itemsets to avoid   making multiple passes over the large original transaction table and   generating a huge sets of candidates.   We have built a parallel database system with DB2 and made   performance evaluation on it. We prove that data mining with   SQL can achieve sufficient performance by the utilization of database   tuning.