Please use this identifier to cite or link to this item: http://repositorio.ufla.br/jspui/handle/1/35637
Title: Utilização de meta-heurísticas para a seleção automática de parâmetros do algoritmo k-segmentos
Other Titles: Use of meta-heuristics for automatic selection of k-segment algorithm parameters
Authors: Ferreira, Danton Diego
Barbosa, Bruno Henrique Groenner
Ferreira, Danton Diego
Barbosa, Bruno Henrique Groenner
Vitor, Giovani Bernardes
Keywords: Curvas principais
K-segmentos
Meta-heurísticas
Otimização baseada em ensino-aprendizagem
Principal curves
K-segments
Metaheuristics
Teaching-learning-based optimization (TLBO)
Issue Date: 24-Jul-2019
Publisher: Universidade Federal de Lavras
Citation: BRAGA, A. de A. Utilização de meta-heurísticas para a seleção automática de parâmetros do algoritmo k-segmentos. 2019. 94 p. Dissertação (Mestrado em Engenharia de Sistemas e Automação)–Universidade Federal de Lavras, Lavras, 2019.
Abstract: Being a nonlinear generalization of principal component analysis, the principal curves technique is a robust tool for data analysis and classification. In pattern recognition one of the most popular algorithms to build Principal Curves is the k-segments algorithm. This algorithm presents good results and excellent applicability due to its guaranteed convergence and robustness. However, its use and performance depend on user-defined parameters. This work presents an automatic selection technique of the quantity and length of segments of the k-segment algorithm with the use of different mono-objective and multiobjective meta-heuristics, especially TLBO (Teaching-Learning-Based Optimization). Different applications of the proposed method are studied, as representation, supervised classification and unsupervised classification of data, for which are used as cost functions equations that take into account the length of the curve and the distance of the events to the segments where they are projected, in addition to minimizing the classification error of the validation bases for the supervised classification of data. Experimental tests made with two-dimensional synthetic and real, mostly multidimensional, databases, taken from known repositories, are presented to demonstrate the efficiency of the proposed method. For representation the quality of results is observed visually, while for supervised and unsupervised classification of data, comparisons are made with the k-NN and k-means methods, respectively. For the supervised classification it is observed that both methods have similar results, highlighting the superiority observed in the proposed method for the database with the largest dimension. For data clustering, it is observed that the proposed method achieves superior results than the comparative method for most databases, depending on the cost-function used, being also observed the importance of multiobjective optimization for this purpose.
URI: http://repositorio.ufla.br/jspui/handle/1/35637
Appears in Collections:Engenharia de Sistemas e automação (Dissertações)



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.