Publication: Técnicas evolutivas multiobjetivo para la evolución de matrices de proyección en problemas de clasificación
Loading...
Identifiers
Publication date
2012-10
Defense date
2012-10-05
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
El trabajo presenta un estudio sobre la utilización de matrices obtenidas con técnicas de computación evolutiva con el fin de realizar proyecciones lineales sobre datos para obtener buenos resultados en el ámbito de la clasificación. Se han realizado distintas experimentaciones, sobre varios dominios de datos, con matrices obtenidas mediante técnicas evolutivas como algoritmos genéticos y estrategias evolutivas, concretamente CMA-ES (Covariance Matrix Adaptation Evolution Strategy). La técnica de clasificación utilizada ha sido k-NN (K nearest neighbors). Además, se han realizado experimentaciones sobre la proyección a dos dimensiones de los datos mediante el uso de matrices evolucionadas y se ha comparado con el uso de algoritmos bien conocidos como PCA y Sammon. Por último, se ha realizado experimentación con algoritmos evolutivos multiobjetivo con el fin de encontrar matrices que proyectasen los datos a la menor dimensión posible aumentando los aciertos en clasificación respecto a los datos sin proyectar. _________________________________________________________________________________________________________________
The paper presents a study on the use of matrices obtained with evolutionary computation techniques to perform linear projections of data to obtain good results in the area of supervised classification. Different experiments have been conducted on various domains of data with matrices obtained by evolutionary techniques like genetic algorithms and evolutionary strategies, specifically CMA-ES (Covariance Matrix Adaptation Evolution Strategy). Supervised classification technique used was k-NN (K nearest neighbors). Furthermore, experiments have been performed on two-dimensional projection using evolved matrices and this method has been compared with the use of well-known algorithms such as PCA and Sammon. Finally, there has been experimenting with multi-objective evolutionary algorithms to find the data matrices that project to the smallest possible dimension increasing hits on supervised classification respect to the data without projection.
The paper presents a study on the use of matrices obtained with evolutionary computation techniques to perform linear projections of data to obtain good results in the area of supervised classification. Different experiments have been conducted on various domains of data with matrices obtained by evolutionary techniques like genetic algorithms and evolutionary strategies, specifically CMA-ES (Covariance Matrix Adaptation Evolution Strategy). Supervised classification technique used was k-NN (K nearest neighbors). Furthermore, experiments have been performed on two-dimensional projection using evolved matrices and this method has been compared with the use of well-known algorithms such as PCA and Sammon. Finally, there has been experimenting with multi-objective evolutionary algorithms to find the data matrices that project to the smallest possible dimension increasing hits on supervised classification respect to the data without projection.
Description
Keywords
Computación evolutiva, Algoritmos genéticos, Clasificación, Matrices