Design And Implementation Of Decision Tree Classifier Based On WEKA

Decision tree classifier takes much importance in the technology of Data mining. As the wide applications of classifier of Data mining, decision tree classifier has achieved much research achievement. Based on deep research in the main algorithms of decision tree, this dissertation finishes the designing and carrying of decision tree classifiers on the platform of WEKA, which not only realizes in effect the utilization of the data miner in existence, but also takes bravery researches and innovations in the rare concerned field of decision tree classifier, and also performs the realizing of new algorithm of decision tree.Firsty, after descriptions in detail of the function and structure of data miner WEKA, based on the research of criterions for evaluating decision tree classifiers'performance, this dissertation takes analytic experiments of classic decision tree classifiers on the platform of WEKA, and does some comparisons and analyses according to different criterions for evaluating.Secondly, this dissertation takes deep researches in the performing theory of classic decision tree classifiers. Based on serious study in the system structure of WEKA, we take the algorithm of SPRINT out, and also take test experiments for the algorithms'performance, which utilizes data miner in effect to carry out individual algorithms.In order to improve the ability of dealing with multi-valued and multi-labled data, this dissertation discusses a new multi-valued and multi-labled data decision tree classifier, on the basis of decision tree classifier in exist. In the new multi-valued and multi-labled decision tree, a new approach of measuring similarity considering both same and consistent features of label-sets is proposed. The new classifier is realized and tested in WEKA, the result of experiment shows that it has better classification efficiency for multi-valued and multi-labled data.