Aplicação de Algoritmos Evolutivos Multiobjetivo na Seleção de Instâncias

Systems for Knowledge Discovery in Databases and Machine Learning predict situations, group and recognize patterns, among other tasks. Although these applications are concerned in generate fast, reliable and easy to interpret information, extensive databases used for such applications make difficult achieving accuracy with a low computational cost. To solve this problem, the databases can be reduced aiming to decrease the processing time and facilitating its storage, as well as, to save only sufficient and relevant information for the knowledge extraction. In this context, methods to reduce and filter databases have been proposed, especially the Instance Selection Methods that selects a subset of examples from the original training data. The subset should maintain all the information of the original set, so that it can be used to generate classification models with the same accuracy as models generated by using the original set. In the last decades, several approaches have been suggested and studied in order to select instances, among them the Evolutionary Algorithms. Although the instance selection methods aim to optimize two goals conflicting with each other, accuracy and reduce computational cost, only an algorithm for multi-objective optimization has been applied to this problem. Therefore, the aim of this study is to perform instance selection based on widely known Multi-objective Evolutionary Algorithms, such as NSGAII and SPEA-II, and to evaluate the classification performance over different domains datasets. The results, compared to others available in the correlative literature, demonstrate that NSGA-II and SPEA-II algorithms can be applied in instance selection for classification problems, because many superfluous instances are removed from the training set, reducing runtime in the classification stages, without significant changes in accuracy