Comparing Sparsification Techniques in the Design of Kernel Prototype -Based Classifiers

Título: Comparing Sparsification Techniques in the Design of Kernel Prototype -Based Classifiers

Autores: David Nascimento Coelho, Guilherme de Alencar Barreto

Resumo: This article addresses the construction of kernelized prototype-based models through sparsification methods. To this end, it extends the algorithm proposed by [1], which employs the ALD method for sparsification and the nearest neighbor classifier based on the selected prototypes. Three additional sparsification methods, namely coherence, novelty, and surprise, are incorporated and compared to the ALD-based approach, along with the use of the weighted K-nearest neighbors algorithm. The results show improvements in the performance of the original algorithm across various datasets, highlighting the strong influence of the chosen sparsification method and kernel function on the classifier’s accuracy rates and the number of selected prototypes.

Palavras-chave: Prototype-based classifiers; sparsification; kernel methods; K-nearest neighbors.

Páginas: 8

Código DOI: 10.21528/CBIC2025-1191412

Artigo em PDF: CBIC_2025_paper1191412.pdf

Arquivo BibTeX:
CBIC_2025_1191412.bib