Restricted Bayesian Network Classification Algorithms Based On The Capability Of Emerging Patterns

Data Mining is the theory and method on researching how to mine knowledge from data in very large databases in nontrivial methods.Classification,as an important theme in data mining,has been researched earlier in statistics,machine learning,neural network,expert systems,etc.In the most of classification models,constructing a classifier using pattern of "attribute -value" pairs is one of important techniques,in which,a classifier built by emerging patterns has been shown better classification accuracy for data sets with strong classified characteristics.But when the attributes of a dataset are not much sensitive to class label,the effectiveness using emerging patterns will get worse.In this paper,we firstly introduce the concept and basic technology about classification.Then we present the basic concept about EPs and efficient mining algorithm of border's operation named MBD-LLBORDER in detail.Also,we briefly expound the idea of classification algorithms whose basic algorithm is EPs-based. Finally,two novel algorithms based EPs are proposed.The first algorithm is a restricted Bayesian network classifier based on the capability of emerging patterns.A novel method is proposed to measure the classification capability of emerging patterns,and the measurement can be further expressed dependency relations in Bayesian network classifier between the attributes.This kind of restricted Bayesian network classifier essentially weakened the assumption of conditional independence in naive Bayes classifier.The other one is a classification using special emerging patterns.We use the ability of emerge pattern to define classification coefficient,and then use it to discover new special emgerge patterns to build classifier.Those new build classifiers mainly focus on measuring the classification coefficient accurately.In order to estimate the accuracy of our algorithms,a series of experiments were conducted on the 20 data sets singled out from the UCI machine learning repository. The experimental results have shown that this algorithm can produce good classification results comparably with other state-of-the-art classification methods such as NB,C4.5, TAN and CAEP.