Far East Journal of Experimental and Theoretical Artificial Intelligence
Volume 4, Issue 1, Pages 1 - 32
(August 2009)
|
|
DIVERSITY GENERATION IN CLASSIFIERS ENSEMBLE
Hamid Parvin (Iran), Mohammad Karami (Iran) and Behrouz Minaei-Bidgoli (Iran)
|
Abstract:
Nowadays, recognition systems are widely used in a variety of applications in almost every field. However, most of classification algorithms exhibit good performance just in specific problems, while due to the lack of enough robustness they perform poorly in other problems. It is unanimously corroborated by the experts of the field that Combination of Multiple Classifiers (CMC) can be utilized as a general solution method for improving the accuracy of classifiers and constructing classifiers with generalization capability. However, CMC proves to be effective only provided that its components are independent or in other words, they produce diverse outputs.
During the past few years many new methods have been proposed for creation of diverse classifiers in an ensemble. This paper proposes four different algorithms for generating diversity in constructing an ensemble of homogeneous classifiers. In the proposed algorithms, the desired diversity is achieved through utilizing three different approaches including using clustering, boosting and adding auxiliary features to the primary dataset. The results of our experiments suggest that the proposed algorithms improve the accuracy of a simple ensemble. |
| Keywords and phrases: diversity generation, classifier fusion, classifier ensembles, clustering, boosting, auxiliary features. |
| Communicated by Shun-Feng Su |
| Number of Downloads: 205 | Number of Views: 590 |
|
