Advanced incremental erasable pattern mining from the time-sensitive data stream

Abstract

Pattern mining has been actively advanced and studied in order to process data that is generated in real time, called incremental data. Erasable pattern mining is a concept that extracts erasable patterns as one of the fields belonging to pattern mining. Erasable patterns have useful information in terms of removal, unlike patterns found by other pattern mining methods. The former erasable pattern mining methods, which target incremental data, regard the importance of data identically. However, in real life, recent data are more important and useful than old data, so erasable pattern mining should also process the latest data more significantly in consideration of the feature of real-world data. For this reason, we propose an erasable pattern mining algorithm using an efficient data structure from the time-sensitive data stream. Since the suggested technique performs erasable pattern mining based on a damped window model, it is possible to relatively prioritize the importance of the recent data over the previous data. In addition, it is possible to quickly reflect and efficiently analyze the newly added incremental data due to a list based data structure of the proposed technique. The results of the various performance tests, which are conducted to compare our algorithm with state-of-the-art algorithms, are analyzed, and the analyses indicate that our approach is more efficient than the other latest approaches.