Chapter 6: Classification Algorithms and Applications

Title: Scaling up the classification accuracy of decision tree classifier for multi-class classification tasks

Author(s): Dewan Md. Farid, Sabbir Arif Siddique, Li Zhang, M.A. Hossain, Chowdhury Mofizur Rahman, Zabih Ghassemlooy

Address: Computational Intelligence Group, Northumbria University, Newcastle upon Tyne, UK | Computational Intelligence Group, Northumbria University, Newcastle upon Tyne, UK | Computational Intelligence Group, Northumbria University, Newcastle upon Tyne, UK | Computational Intelligence Group, Northumbria University, Newcastle upon Tyne, UK | Department of Computer Science and Engineering, United International University, Bangladesh; Northumbria Communications Research RLab and Optical Communications Research Group, Northumbria University, Newcastle upon Tyne, UK

Reference: Software, Knowledge, Information Management and Applications (SKIMA 2013) pp. 262 - 272

Abstract/Summary: A decision tree classifier is a useful and efficient data mining method for solving the classification problems in supervised learning. In this paper, we introduce a learning algorithm to improve the classification accuracy rates of decision tree classifier for the classification of multi-class problems. The proposed approach initializes weights to the training instances using naïve Bayes classifier, and removes the training instances from the training data having weights less than 0.5. As the presence of these instances may cause the generated decision tree to suffer from over fitting and decreased accuracy, we remove the noisy troublesome training instances from the training data before the decision tree induction. We tested the performance of the proposed algorithm against the traditional C4.5 decision tree classifier using the classification accuracy, precision, sensitivity-specificity analysis, and ten-fold cross-validation on real benchmark datasets from University of California, Irvine (UCI) machine learning repository. The experimental analysis shows that the proposed algorithm has produced impressive results in the real life challenging multi-class classification problems.

Order a copy of this article Order a copy of this article