Attribute Oriented Induction of High-level Emerging Patterns

Attribute Oriented Induction (AOI) produces high-level characteristic summary data but does not discover new emerging patterns. Emerging Pattern (EP) algorithms discover emerging patterns between datasets but mostly consider low-level data. This paper introduces an algorithm, AOI-HEP, derived from both AOI and High-level Emerging Patterns (HEP), where HEP discriminates the high level data from AOI. The main objective is to discover characteristic HEP patterns using AOI. To filter out the large overlapping and subsuming attribute values in the output, a Cartesian product of attribute values, a similarity metric based on attribute values and attribute hierarchy level are applied. Experiments used four datasets from the UCI machine learning repository. Results show that various interesting HEP patterns can be generated by using the AOI-HEP algorithm.