Clustering Based Multidimensional Sequential Pattern Mining

Abstract

Sequential pattern mining is extracting frequent subsequences as patterns from a sequence database. Sequential pattern mining is an important data mining task and has a variety of applications like analyzing the customer purchasing behaviors, mining the user access patterns for the websites, prediction of natural disasters, using the history of symptoms to predict certain kind of diseases, analysis of DNA sequences. The patterns found from sequential pattern mining, though discover global regularity among customers, may suffer from a lack of focus. That is because sequence patterns are usually associated with different circumstances and such circumstances form a multi dimensional space. Multidimensional sequential patterns, which associate sequential patterns with multiple dimensions, are interesting and useful in practice since people are often interested in detailed sequential patterns associated with different circumstances. A new algorithm ClusterUniSeq is proposed which first performs clustering on the multidimensional sequence database and then mines multidimensional sequential patterns from the clustered multidimensional sequence database. The proposed algorithm improves the quality of extracted patterns since it conveys more information than the existing UniSeq algorithm.