An Efficient Single-Scan Algorithm for Mining Essential Jumping Emerging Patterns for Classification

Emerging patterns (EPs), namely itemsets whose supports change significantly from one class to another, were recently proposed to capture multi-attribute contrasts between data classes, or trends over time. Previous studies show that EP/JEP(jumping emerging patterns) - based classifiers such as CAEP[2] and JEP-classifier[6] have good overall predictive accuracy. But they suffer from the huge number of mined EPs/JEPs, which makes the classifiers complex.In this study, we propose a special type of EP, essential jumping emerging patterns (eJEPs), which are believed to be high quality patterns with the most differentiating power and thus are sufficient for building accurate classifiers. Existing algorithms such as border-based algorithms and consEPMiner[7] can not directly mine such eJEPs. We present a new single-scan algorithm to effectively mine eJEPs of both data classes (both directions). Experimental results show that the classifier based exclusively on eJEPs, which uses much fewer JEPs than JEP-classifier, achieves the same or higher testing accuracy and is often also superior to other state-of-the-art classification systems such as C4.5a nd CBA.