Title: Evolving novel classification measures, optimising and evaluating the weights in weighted average random forest using quadratic programming

Authors: Kala Nisha Gopinathan; Punniyamoorthy Murugesan; M. Priyadharsan; L.J. Dharanidharan

Addresses: Department of Management Studies, National Institute of Technology, Tiruchirappalli, Tamil Nadu, 620015, India ' Department of Management Studies, National Institute of Technology, Tiruchirappalli, Tamil Nadu, 620015, India ' National Institute of Technology, Tiruchirappalli, Tamil Nadu, 620015, India ' National Institute of Technology, Tiruchirappalli, Tamil Nadu, 620015, India

Abstract: Attribute selection measures are used in decision trees to select the feature that best splits the data into homogeneous parts. There are four existing measures: the Gini index, entropy, information gain, and gain ratio. In this paper, two novel attribute measures were proposed and were tested on 10 different datasets using decision trees to find their effectiveness. Statistical tests conducted on the proposed novel attribute selection measures showed that they were able to achieve the same level of performance as the existing measures, while eliminating the limitations in the existing measures. In the weighted average random forest, the weights were optimised by minimising the objective function of mean squared error (MSE). This has been done through Quadratic Programming, and the computed optimal weights were used with the decision trees in the weighted average random forest. Overall, it was determined that the weighted average random forest outperformed the random forest.

Keywords: Gini impurity; entropy; information gain; gain ratio; new gain ratio; purity index.

DOI: 10.1504/IJENM.2025.151291

International Journal of Enterprise Network Management, 2025 Vol.16 No.4, pp.339 - 366

Received: 09 May 2024
Accepted: 22 Aug 2024
Published online: 22 Jan 2026 *

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