| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 102 === Classification is always an important part in data mining. The information of data can help us predicting what will happen in the future. Therefore, there are different merits for different classifiers to process various classified problems. Among many classifiers, Naive Bayes Classifier works more effectively and stably, so it is the most popular classifier in practical applications.
In this paper, firstly, the Decision Tree based on Bayes Classifier is used to decide the major attributes that are considered to give more contribution in terms of classification. Secondly, the initial weight for each of the selected major attributes is set up based on the nature of Decision Tree in which the level distance from the tree root reflects the role of importance to the classification. Thirdly, the weight of each selected major features is then adjusted by the least mean square learning process. Finally, the resultant weights of major features from the learning are used in the Bayes Classifier for classification and prediction. The experiment shows that the recognizing ability of the classifier proposed in this thesis is relatively better against other some compared approaches for various Datasets in the different domains. The experimental result also shows the proposed approach has better robustness.
|
|---|
