الفهرس 
AcknowledgmentOnly 14 pages are availabe for public view
 |  | from | 202 |  |  |
| | | from | 202 | | |
Abstract
Classification based on patterns is a relatively new methodology. The previously proposed classifiers of EPs family are EP classifier, Jumping Emerging Patterns classifier (JEP) and Strong Jumping Emerging Patterns classifier (SJEP). All of these classifiers suffer from problems with its accuracy and the memory consumption during the mining process. The proposed Modified Strong Jumping Emerging Patterns (MSJEPs) are those item sets whose support increases from zero in one data set to non-zero in the other dataset with support constraint greater than minimum support threshold (ζ).The support constraint of MSJEP removes potentially less useful JEPs while retaining those with high discriminating power. The proposed MSJEP solves the weakness point in SJEP classifier that constructs one Contrast Pattern tree (CP-tree) for the whole data set to generate SJEPs. CP-tree-based discovery algorithm used for SJEP mining is a main-memory-based method. When the data set is large, it is unrealistic to assume that the CP-tree can fit in main memory. The main idea to handle this problem is to first partition the data set into a set of projected data sets and then for each projected data set, we construct and mine its corresponding CP-tree. Trees of the projected data sets are called Separated Contrast Pattern Tree (SCP-trees) and Patterns generated from it are Called MSJEPs. The proposed algorithm has the same superior accuracy of SJEP classifier but solves the problem of memory consumption. The proposed MSJEP classifier can solve another weakness point of EPs, JEPs and SJEPs because these classifiers are weak in handling attributes whose values are associated with taxonomies. MSJEP classifier achieves better accuracy and better speed than those classifiers and also is the best in handling attributes in taxonomy. Experiments on a large number of benchmark data sets show that the MSJEP Classifier is consistent, highly effective at classifying various kinds of datasets, and usually achieves higher accuracy than other state-of-the-art classifiers. The proposed MSJEP classifier solves the problem of memory consumption in SJEP and has the same superior results. Also, G.MSJEP is handling attributes associated with taxonomies.