Search

Mining frequent patterns in transaction databases has been a popular subject in data mining research. Common activities include finding patterns in database transactions, times-series, and exceptions. The Apriori algorithm is a widely accepted method of generating frequent patterns. The algorithm can require many scans of the database and can seriously tax resources. New methods of finding association rules, such as the Frequent Pattern Tree (FP-Tree) have improved performance, but still have problems when new data becomes available and require two scans of the database.This paper proposes a new method, which requires only one scan of the database and supports update of patterns when new data becomes available. We design a new structure called Pattern Repository (PR), which stores all of the relevant information in a highly compact form and allows direct derivation of the FP-Tree and association rules quickly with a minimum of resources. In addition, it supports run-time generation of association rules by considering only those patterns that meet on-line data requirements.