Incremental Sequential Pattern Tree Mining

Abstract

Sequential pattern mining, which discovers the correlation relationships from the ordered list of events, is an important research field in data mining area. In our study, we have developed a Sequential Pattern Tree structure to store both frequent and non-frequent items from sequence database. It requires only one scan of database to build the tree due to storage of non-frequent items which reduce the tree construction time considerably. Then, we have proposed an efficient Sequential Pattern Tree Mining algorithm which can generate frequent sequential patterns from the Sequential Pattern Tree recursively. The main advantage of this algorithm is to mine the complete set of frequent sequential patterns from the Sequential Pattern Tree without generating any intermediate projected tree. Again, it does not generate unnecessary candidate sequences and not require repeated scanning of the original database. We have compared our proposed approach with three existing algorithms and our performance study shows that, our algorithm is much faster than apriori based GSP algorithm and also faster than existing PrefixSpan and Tree Based Mining algorithm which are based on pattern growth approaches.