Finding sequential patterns with TF-IDF metrics in health-care databases

Abstract

Finding frequent sequential patterns has been defined as finding ordered list of items that occur more times in a database than a user defined threshold. For big and dense databases that contain really long sequences and large itemset such as medical case histories, algorithm proposed on this idea of counting the occurrences output enourmous number of highly redundant frequent sequences, and are therefore simply impractical. Therefore, there is a need for algorithm that perform frequent pattern search and prefiltering simultaneously. In this paper, we propose an algorithm that reinterprets the term support on text mining basis. Experiments show that our method not only eliminates redundancy among the output sequences, but it scales much better with huge input data sizes. We apply our algorithm for mining medical databases: what diagnoses are likely to lead to a certain future health condition.