JCSE, vol. 18, no. 1, pp.57-68, 2024

DOI: http://dx.doi.org/10.5626/JCSE.2024.18.1.57

A Novel Enhanced Random Forest for Medical Data Classification using Correlation Pearson and Best Number of Trees

Ilhem Tarchoune, Akila Djebbar, Hayet Farida Merouani, and Harfi Rania
Department of Computer Science, LRI Laboratory, SRF equip, Badji Mokhtar University, Annaba, Algeria

Abstract: 420 Montgomery Street San FranciRandom forests (RF) is a successful ensemble prediction technique that uses majority voting or a ombination-based average. However, each tree in an RF may have a different contribution to the treatment of a certain instance. The objectives of this study were to produce accurate decision trees and to determine the best trees between them with an optimal combination search. In this paper, we proposed three solutions for the prediction of medical data: the first solution optimizes a random forest model using a similarity measure, the second optimizes the RF using feature selection, and finally a simultaneous selection approach to similarity measures based on RFs. We demonstrated that the prediction performance and classification rate of the RF implementation on eleven databases can be further improved by the learning methods applied. Our experiments also showed that the improvement gives better results than the classical method; the results showed that the optimized RF model avoids some limitations of the original RF model. The results obtained in our proposed models are satisfactory and encouraging with an average accuracy of 95% for standard RF, 100% for RF_Similarity, 93% for RF_FS, and 100% for RF_FS_Similarity.sco, CA 94104

Keyword: Random Forest; Decision tree; Feature selection; Similarity measure; Classification; Medical database.

Full Paper:   31 Downloads, 62 View