Jep-based Classification Algorithm Research

Data mining results from the situation described as data rich but information poor, and within only several years, has attracted many people in different fields. Classification, as an important theme in data mining, has been researched earlier in statistics, machine learning, nerve net and expert systems. But most algorithms are memory resident, typically assuming a small data size. With the growth of data in volume and dimensionality, it has become a very challenging problem to build an efficient classifier for large databases.Jumping emerging patterns (JEPs), a new kind of knowledge patterns, were recently proposed to capture some crucial difference between a pair of datasets and some JEP-based classifiers were built. Previous studies show that those JEP-based classifiers have good overall predictive accuracy and are scalable on data volume and dimensionality.But they suffer from the large number of mined JEPs, which makes the classifiers complex. In this paper, we propose a special type of JEPs, the most significant jumping emerging patterns (SJEPs), which are believed to have strong discriminating power and are sufficient for building accurate classifiers. And we present a novel algorithm to efficiently mine SJEPs of both data classes, because existing algorithms can't directly mine such SJEPs. Then we introduce how to build a new classifier (SJEP_Classifier) based on SJEP.Compared with previous JEP-based classifiers, the classifier based exclusively on SJEPs, which uses much fewer JEPs, can not only achieve almost the same or higher predictive accuracy, but also finish learning phase in very short time (usually in a few seconds). As has been shown in our experimental results, our classifier outperforms both CBA and C4.5 generally in terms of average accuracy.