A Diversity-Accuracy Measure for Homogenous Ensemble Selection

TitleA Diversity-Accuracy Measure for Homogenous Ensemble Selection
Publication TypeJournal Article
Year of PublicationIn Press
AuthorsZouggar, T. S., and A. Adla
JournalInternational Journal of Interactive Multimedia and Artificial Intelligence
IssueIn Press
VolumeIn Press
NumberIn Press
Date Published06/2018

Several selection methods in the literature are essentially based on an evaluation function that determines whether a model M contributes positively to boost the performances of the whole ensemble. In this paper, we propose a method called DIversity and ACcuracy for Ensemble Selection (DIACES) using an evaluation function based on both diversity and accuracy. The method is applied on homogenous ensembles composed of C4.5 decision trees and based on a hill climbing strategy. This allows selecting ensembles with the best compromise between maximum diversity and minimum error rate. Comparative studies show that in most cases the proposed method generates reduced size ensembles with better performances than usual ensemble simplification methods.

KeywordsBagging, Classification, Decision Trees, Ensemble Methods, Ensemble Pruning, Hill Climbing, Machine Learning
ip_06_05.pdf755.31 KB