Ensembles with Clustering-and-Selection Model Using Evolutionary Algorithms

Abstract

Ensembles of classifiers is a way to improve the performance of the approach with single classifiers. The idea is to find and combine a set of classifiers that are responsible for smaller and theoretically easier parts of a problem to solve, in other words, divide to conquer. Between the ensembles models, there is the clustering and selection in which the training data are clustering, and a classifier is built for each cluster found. An answer for an input data is given based on a distance to the available clusters that has an associated classifier. In this paper, the clustering and selection model is explored with the use of Evolutionary Algorithms to search clusters that optimize the ensemble's performance. Experiments are conducted with ten datasets and using recent advances in classification methods. The results achieved good and promising performances compared to classical clustering-and-selection model and other methods to build ensembles.