Banca de DEFESA: GUSTAVO TORRES CUSTODIO

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
DISCENTE : GUSTAVO TORRES CUSTODIO
Data: 07/10/2026
HORA: 09:00
LOCAL: https://conferenciaweb.rnp.br/sala/debora-149
TÍTULO:

Class-Separation-Guided Ensemble Construction with Evolutionary Classifier Selection for Improved Classification

PÁGINAS: 135
RESUMO:

Classification problems are common in Machine Learning, with applications ranging from spam detection to credit analysis. When individual classifiers are unable to adequately separate classes, ensemble methods can combine multiple classifiers to improve predictive performance. In this work, we propose CBEG (Clustering-Based Ensemble Guidance), a framework that combines data clustering, feature selection, base classifier selection, and classifier fusion to construct specialized ensembles. In CBEG, classifiers are trained using data associated with different clusters, where the clustering process is designed to increase the separation between
classes within each cluster. The framework introduces DBC (Distance Between Classes), a metric for selecting the optimal cluster partition based on the distance between different classes, and a classifier voting strategy based on cluster membership degrees to determine the predicted class of each sample. CBEG was evaluated
through an ablation study and a comparative analysis involving binary-class, multiclass, and synthetic datasets. The ablation study showed that the contribution of individual components varies across datasets, with base classifier selection, particularly PSO-based selection, providing the most consistent improvements. Feature selection generally low impact on the overall performance, while the effectiveness of the fusion strategies depended on the dataset and configuration. CBEG achieved notable performance when compared with other ensemble methods, obtaining particularly strong results with the Pima Diabetes and Blood Transfusion datasets. On the synthetic datasets, CBEG achieved its highest F1-scores on datasets generated
from normal distributions. The proposed DBC metric and the Adjusted Rand Index (ARI), employed for clustering evaluation, were compared. DBC achieved the best results more frequently for accuracy and recall, whereas ARI performed best more frequently for precision, F1-score, and AUC-ROC. Regarding classifier fusion, the proposed cluster membership strategy was outperformed by the meta-classifier. Finally, a literature review was conducted to examine existing approaches for improving classifier ensembles through the combination of evolutionary algorithms and data clustering. Overall, the results demonstrate that CBEG can effectively leverage clustering and classifier selection to construct specialized ensembles.

 

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 1918407 - DEBORA MARIA ROSSI DE MEDEIROS
Membro Titular - Examinador(a) Interno ao Programa - 1673092 - RONALDO CRISTIANO PRATI
Membro Titular - Examinador(a) Interno ao Programa - 1934625 - JESUS PASCUAL MENA CHALCO
Membro Titular - Examinador(a) Externo à Instituição - ANA CAROLINA LORENA - ITA
Membro Titular - Examinador(a) Externo à Instituição - CHARLES HENRIQUE PORTO FERREIRA - FEI
Membro Suplente - Examinador(a) Interno ao Programa - 3008222 - PAULO HENRIQUE PISANI
Membro Suplente - Examinador(a) Externo à Instituição - ADRIANA CAMARGO DE BRITO - IPT-SP
Membro Suplente - Examinador(a) Externo à Instituição - MARCIO BASGALUPP - UNIFESP