A Patricia-Trie Approach for Incremental Mining of Frequent Itemsets on Indexed Data Blocks

Patricia-Trie based I-Forest index is a novel index structure that supports efficient item set mining into a relational DBMS. The The Patricia-Trie based I-Forest index provides a complete and compact representation of transactional data. It is a general structure that efficiently supports different algorithmic approaches to item set extraction. Selective access of the physical index blocks significantly reduces the I/O costs and efficiently exploits DBMS buffer management strategies. This approach, albeit implemented into a relational DBMS, yields performance better than the state-of-the-art algorithms accessing data on a flat file and is characterized by a linear scalability also for large data sets.