Title: An ensemble framework of decision trees for class imbalance using partitioning

Authors: Vijayakumar Kadappa; Shankru Guggari; Rajeshwari Devi Doddapoojari Veerabhaskar

Addresses: Department of Computer Applications, B.M.S. College of Engineering, Bengaluru 560019, India ' School of Computer Science and Engineering, KLE Technological University, Hubballi 580031, India ' Department of Electronics and Communication Engineering, Global Academy of Technology, Bengaluru 560098, India

Abstract: Decision tree classifiers are widely used in machine learning and data mining due to their intuitiveness. However, they do not perform well for class-imbalanced data due to bias creation towards the majority class. Therefore, handling class-imbalanced data is an active research area in many applications of machine learning. Feature set partitioning paradigm is proven to be effective for classification by many researchers. In this paper, we extend the ideas of partitioning to propose a decision tree ensemble to overcome class-imbalance issues. The framework consists of balancing the data, partitioning the feature set, building local decision trees, then combining decisions. Some instances of the proposed framework (e.g., Ferrer diagram-based approach) are used to validate the performance of the framework. An empirical study is carried out using University of California Irvine (UCI), Knowledge Extraction Evolutionary Learning (KEEL), and other datasets. The performance of the instances of the proposed framework demonstrates improved performance compared to other benchmark machine learning approaches.

Keywords: machine learning; decision tree; feature set partitioning; class imbalance; ensemble method; classification.

DOI: 10.1504/IJICA.2024.143393

International Journal of Innovative Computing and Applications, 2024 Vol.15 No.1, pp.1 - 13

Received: 12 Jul 2023
Accepted: 17 Feb 2024
Published online: 17 Dec 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article