Experimental evaluation of diversity measures of combined classifiers

Authors:
Fuad Alkoot;Husain Qasem
Affiliations:
Telecommunication and Navigation Institute, PAAET, Kuwait;Telecommunication and Navigation Institute, PAAET, Kuwait
Venue:
AIKED'10 Proceedings of the 9th WSEAS international conference on Artificial intelligence, knowledge engineering and data bases
Year:
2010

Citing 8
Cited 0

Decision Combination in Multiple Classifier Systems

IEEE Transactions on Pattern Analysis and Machine Intelligence
Hierarchical mixtures of experts and the EM algorithm

Neural Computation
Bagging predictors

Machine Learning
On Combining Classifiers

IEEE Transactions on Pattern Analysis and Machine Intelligence
The Random Subspace Method for Constructing Decision Forests

IEEE Transactions on Pattern Analysis and Machine Intelligence
An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization

Machine Learning
Modified product fusion

Pattern Recognition Letters
Combining Pattern Classifiers: Methods and Algorithms

Combining Pattern Classifiers: Methods and Algorithms

Quantified Score

Hi-index	0.00

Visualization

Abstract

We aim to find the effect of diversity on combiner performance. Three diversity measures are used to calculate the diversity of combined classifiers. We aim to identify which measure is closely related to the combiner performance. Three combiner types are used; Bagging and a conventional three classifier system, in which three classifier types are used; backpropagation neural network, bayesian and k-nearest neighbor classifiers. Additionally we experiment with a feature based combiner system proposed by Alkoot [13,14]. Results obtained on real data indicate the diversity measure to be higher for systems with higher classification rate, if it outperforms other classifiers by a large margin. Otherwise, if the performances of the compared systems are close the diversity measure may not be higher for the best system. On many occasions the diversity measures were not good indicators of system performance. On some instances we found that the more diverse system did not yield a better performance.