Clustering-based dynamic ensemble selection for one-class decomposition

• Ano de defesa: 2022
• Autor(a) principal: FRAGOSO, Rogério César Peixoto
• Orientador(a): CAVALCANTI, George Darmiton da Cunha
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Pernambuco
• Programa de Pós-Graduação: Programa de Pos Graduacao em Ciencia da Computacao
• País: Brasil
• Palavras-chave em Português: Inteligência computacional; Sistemas de múltiplos classificadores
• Link de acesso: https://repositorio.ufpe.br/handle/123456789/48095

A natural solution to tackle multi-class problems is employing multi-class classifiers. How- ever, in specific situations, such as imbalanced data or a high number of classes, it is more effective to decompose the multi-class problem into several and easier to solve problems. One- class decomposition is an alternative, where one-class classifiers (OCCs) are trained for each class separately. However, fitting the data optimally is a challenge for classifiers, especially when it presents a complex intra-class distribution. The literature shows that multiple classifier systems are inherently robust in such cases. Thus, the adoption of multiple OCCs foreach class can lead to an improvement for the one-class decomposition. With that in mind, in this work, we introduce two methods for multi-class classification using ensembles of OCCs. One-class Classifier Dynamic Ensemble Selection for Multi-class problems (MODES, for short) and Density-Based Dynamic Ensemble Selection (DBDES) provide competent classifiers for each region of the feature space by decomposing the original multi-class problem into multiple one-class problems, segmenting the data from each class, and training a OCC for each cluster. The rationale is to reduce the complexity of the classification task by defining a region of the feature space where the classifier is supposed to be an expert. The classification of a test instance is performed by dynamically selecting an ensemble of competent OCCs and the final decision is given by the reconstruction of the original multi-class problem. Experiments carried out with 25 databases, 4 OCC models, and 3 aggregation methods showed that the proposed techniques outperform the literature. When compared with literature techniques, MODES and DBDES obtained better results, especially for databases with complex decision regions.