Rare Association Rules Mining of Diabetic Complications Based on Improved Rarity Algorithm

Abstract

Although the frequent pattern mining has attracted widespread attention of scholars, it is undeniable that the rare pattern mining plays a significant role in many fields, such as medical, financial, and scientific fields. And it is more valuable to study the rare pattern mining, because it tends to find some unknown and unexpected associations. There are some previous algorithms of rare itemsets mining, however, Arima spends much time and Rarity wastes much space. So based on the Rarity algorithm, this paper presents an improved top-down approach to efficiently mine all rare itemsets and their association rules, which uses the graph structure to indicate all combinations of existing items in the database, defines a pattern matrix to record all itemsets and the support_count, and combines the hash table to accelerate support calculation to quickly find all rare itemsets, and then generate all patterns to choose useful rules according to their interesting rate. In the experiment, this paper uses the real diabetic clinical data to verify this improved approach and mines some useful rules among the diabetic complications. Moreover, compared with the two methods mentioned above, this method decreases much time and space complexity in the association rules mining.