The Integrated Classification Algorithm Via Emerging Patterns Based On Boosting Technology

With the fast development of the data size and data dimensions, the concept ofdata mining was proposed by many researchers, which is a kind of technology and canbe used to mine and extract some useful knowledge patterns from a large amount ofdata. As an important analysis and process method to data in the past years,classification technology was a significant research topic in the scientific fields ofmachine learning, pattern recognition, and statistics. Now, it is also a key task of datamining. Some researchers proposed many classification methods because of its broadprospect for applications. The current research results show that, every classificationmethod has its own advantages and disadvantages. In addition to this, it is alsopointed that, all classification algorithms are not isolated, and they can learn fromeach other. As a matter of fact, the excellent classification solution in practicalapplications is always the compositive result of various classification methods.Based on the above conclusions, we proposed to integrate some classifiers withan integration learning method, so as to improve their classification accuracy.Focusing on this point, a simple and intuitive knowledge pattern, namely emergingpattern, is studied deeply by us. Emerging patterns are those elements whose supportrate is changed dramatically from a data set to another. The emerging pattern cancapture the differences of a group of attributes between the target class and thenon-target ones, in other words, it has a good distinguish ability. Because of the aboveadvantages of the emerging patterns, we decide to construct the basic classifiers of theintegrated classification algorithm proposed in this paper based on them.The boosting method is chosen by us to integrate the basic classifiers constructedin the last stage. We propose an integrated classification algorithm via emergingpatterns based on boosting technology in this paper. The test results on the givenstandard databases show that, the performance of the integrated classificationalgorithm via emerging patterns based on boosting technology proposed in this paper,as a whole, is superior to some existing good algorithms, such as C4.5, CBA, CAEP,and NB. In addition to this, we demonstrated that, the classification precision of theintegrated algorithm via emerging patterns based on boosting method is higher thanthe one based on bagging technology. At last, we get the conclusion that theperformance of a simple classifier can be improved by integrating various basic classifiers based on the boosting technology.