Research On Bayesian Classifier For Ranking

Classification is one of the most important issues in data mining and machine learning. Bayesian algorithms have a solid theoretical foundation of mathematics, So they are efficient on classifications and have been important algorithms of the Classification algorithms. However, in practical application of data mining and machine learning, only one accurate classification algorithm on data is not enough. An accurate Ranking algorithm is often needed, and it is even more important than a accurate classification algorithm. Therefore, the Ranking research based on bayes is interesting and is of great significance.We build an Experimental Platform on Weka framework. On this platform, we implement Naive Bayes (NB), Tree Augmented Naive Bayes (TAN), NBTree, Hidden Naive Bayes (HNB) and Aggregating One-Dependence Estimators (AODE) classifier algorithms. By the contrast with these classifications we found, the results of AODE classifier on some data set are not as good as we expected by the reason that we didn't fully consider the importantance of the relations between non-class attributes and class attribute. By the weighted values on the classifier of AODE, WAODE classifier overcomes deficiencies and improves the efficiency of the classification. So, we try to apply the WAODE algorithm on Ranking.On the Ranking platform, we do experiments on six algorithms above. The results show that on 35 UCI data set WAODE classifier algorithm( the expand classifier algorithm of AODE ) is better than AODE classifier algorithm especially on big data sets. And the WAODE algorithm is the best in the six algorithms above.