Extending SWF for Incremental Association Mining by Incorporating Previously Discovered Information

摘要

Incremental mining of association rules from dynamic databases refers to the maintenanceand utilization of the knowledge discovered in the previous mining operations.Sliding-window-filtering (SWF)is a technique proposed to filtering false candidate 2-itemsets bysegmenting a transaction database into several partitions.SWF computes a set of candidate2-itemsets that is close to frequent 2-itemsets.Therefore,it is possible to generate several candidate k -itemsets for one database scan.Such a database scan reduction technique greatly increase the performance for frequent itemsets discovery.In this paper,we extend SWF by incorporating previously discovered information and propose two algorithms to boost theperformance for incremental mining.The first algorithm FI SWF (SWF with FrequentItemset)reuse the frequent itemsets (and the counts)of previous mining task as FUP2 toreduce the number of new candidate itemsets that have to be checked.The second algorithmCI SWF (SWF with Candidate Itemset)reuse the candidate itemsets (and the counts)from the previously mining task. Experimental studies are performed to evaluate performance of the new algorithms. The study shows that the new incremental algorithm is significantly faster than SWF. More importantly, the need for more disk space to store the previously discovered knowledge does not increase the maximum memory required during the execution time.