Academic Journals Database
Disseminating quality controlled scientific knowledge

Scaling Up the Accuracy of Decision-Tree Classifiers: A Naive-Bayes Combination

ADD TO MY LIST
 
Author(s): Liangxiao Jiang | Chaoqun Li

Journal: Journal of Computers
ISSN 1796-203X

Volume: 6;
Issue: 7;
Start page: 1325;
Date: 2011;
Original page

Keywords: naive Bayes | decision trees | class-membership probabilities | weights | classification | ranking

ABSTRACT
C4.5 and NB are two of the top 10 algorithms in data mining thanks to their simplicity, effectiveness, and efficiency. In order to integrate their advantages, NBTree builds a naive Bayes classifier on each leaf node of the built decision tree. NBTree significantly outperforms C4.5 and NB in terms of classification accuracy. However, it incurs very high time complexity. In this paper, we propose a very simple, effective, and efficient algorithm based on C4.5 and NB. We simply denote it C4.5-NB. Our motivation is to keep the high classification accuracy of NBTree without incurring the high time complexity. In C4.5-NB, C4.5 and NB are built and evaluated independently at the training time, and the class-membership probabilities are weightily averaged according to their classification accuracies on training data at the test time. Empirical studies on a large number of UCI data sets show that it performs as well as NBTree in terms of classification accuracy, but is significantly more efficient than NBTree.
Why do you need a reservation system?      Save time & money - Smart Internet Solutions