A Novel Approach in Mining Specialized Coherent Rules in a Level-Crossing Hierarchy

Abstract

Multi-level association rule mining techniques that are developed for mining data at multiple-level taxonomy usually rely on minimum support and minimum confidence threshold for extracting frequent items at various levels. But the usage of above threshold alone does not produce meaningful and interesting rules. Moreover, the determination of minimum support and confidence is very hard in a dataset containing multiple-level taxonomy. In order to overcome the above drawbacks, coherent rule mining was proposed. The proposed approach first mines multiple-level coherent rules in a predefined taxonomy and then extracts specialized coherent rules. The rules are mentioned as specialized rules because they exist only in the bottommost level of taxonomy irrespective of their predecessor rules. The proposed algorithm works for the quantitative dataset, and a triangular membership function is used for converting the quantitative dataset at various levels into fuzzy itemset. The results obtained are compared with multi-level fuzzy association rule mining technique.