Abstract: In this study, commonly used proximity measures, in other words, distance and similarity functions are reviewed. An interactive software tool with graphical user interface is developed to examine the effect of proximity measures on the performance of clustering methods. The software named ClustProX allows the users to select a data set, to apply a clustering method and to observe the performance of the results. Xie-Beni and Kwon indices are used to evaluate the performance of the implemented clustering methods with different proximity measures and the number of clusters. In this way, the users will be able to choose an appropriate proximity function and the number of clusters to improve accuracy of clustering.

• Publication Date: 03-Aug-2014
• DOI: 10.15224/978-1-63248-005-7-96
• Views: 0
• Downloads: 0

Abstract: Pruning an ensemble of classifiers is one of the most significant and effective issues in ensemble method topic. This paper presents a new ensemble pruning method inspired by upward stochastic walking idea. Our proposed method incorporates simulated annealing algorithm and forward selection method for selecting models through the ensemble according to the probabilistic steps. Experimental comparisons of the proposed method versus similar ensemble pruning methods on a heterogeneous ensemble of classifiers demonstrate that it leads to better predictive performance and small-sized pruned ensemble. One of the reasons of these promising results is more time which our method spends for finding the best models of ensemble compared with rivals

• Publication Date: 03-Aug-2014
• DOI: 10.15224/978-1-63248-005-7-97
• Views: 0
• Downloads: 0