Selección de instancias para el clasificador asociativo gamma usando la teoría extendida de los conjuntos aproximados
Fecha
2016-05-25
Autores
Antón Vargas, Jarvin Alberto
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
El reconocimiento de patrones constituye una ciencia multidisciplinaria que se ha
caracterizado por un continuo desarrollo en disímiles aristas. El enfoque asociativo
en los últimos años se ha caracterizado por crear métodos de una marcada sencillez
y eficacia, comparable en la mayoría de los casos con muchos de los modelos
desarrollados siguiendo otros enfoques, aunque no está exento de sufrir las mismas
deficiencias que estos otros métodos.
En el presente trabajo se realiza una propuesta de selección de instancias para el
clasificador asociativo Gamma desarrollado por López Yáñez en el 2007, en el Centro
de Investigación en Computación del Instituto Politécnico Nacional de México,
basado en la teoría extendida de los Conjuntos Aproximados (Rough Sets), tomando
como base elementos del propio clasificador, logrando eliminar las instancias que
pueden afectar la precisión del mismo. De esta forma se propone el uso de la teoría
extendida de los Conjuntos Aproximados, lo que constituye la base principal del
método de selección de instancias GRIS (Gamma Rough based Instance Selection).
Para la validación del nuevo enfoque, se tomaron como medidas, la exactitud
lograda por el clasificador asociativo Gamma sobre bases de datos balanceadas, el
promedio de la exactitud de la clasificación por clases sobre bases de datos desbalanceadas
y el índice de retención de objetos, y se evalúa con otros modelos de
selección de instancias como el ENN, GGE, RNGE, MSEditB, basados en el error
de clasificación, por lo que se consideran como de propósito similar. La evaluación se
realizó sobre un conjunto de bases de datos internacionales. De esta forma los resultados
del nuevo método presentaron similar comportamiento con respecto a los otros
modelos con respecto a la precisión de la clasificación. De igual forma se obtuvo una
mayor reducción de la cantidad de instancias en la mayoría de las bases de datos.
Pattern recognition is a multidisciplinary science that has been characterized by a continuous development in dissimilar edges. The associative approach in recent years has been characterized by creating methods with a marked simplicity and efficiency, comparable in most cases with many of the models developed in accordance with other approaches, although not exempt from suffering the same deficiencies as these other methods. In this paper a proposal for instance selection for associative Gamma classifier developed by López Yáñez in 2007, at the Center for Computing Research, National Polytechnic Institute of Mexico, based on the extended theory of rough sets is made, using elements of the classifier itself, achieving eliminate instances that may affect the accuracy of the same. Thus the use of the extended theory of rough sets is proposed to be the main basis of the instance selection method GRIS (Gamma Rough based Instance Selection). To validate the new approach were taken as measures, the accuracy achieved by the associative classifier Gamma, the average of the accuracy and Objects Retention Index, evaluated with other models from the state of the art such as the ENN, GGE, RNGE, MSEditB, all these methods based at classification error, which are considered of similar purpose. The evaluation was conducted on a set of international databases. Thus the results of the new method presented similar behavior with respect to the other models regarding the classification accuracy. By other hand the reduction of the number of instances obtained was superior in most of the databases.
Pattern recognition is a multidisciplinary science that has been characterized by a continuous development in dissimilar edges. The associative approach in recent years has been characterized by creating methods with a marked simplicity and efficiency, comparable in most cases with many of the models developed in accordance with other approaches, although not exempt from suffering the same deficiencies as these other methods. In this paper a proposal for instance selection for associative Gamma classifier developed by López Yáñez in 2007, at the Center for Computing Research, National Polytechnic Institute of Mexico, based on the extended theory of rough sets is made, using elements of the classifier itself, achieving eliminate instances that may affect the accuracy of the same. Thus the use of the extended theory of rough sets is proposed to be the main basis of the instance selection method GRIS (Gamma Rough based Instance Selection). To validate the new approach were taken as measures, the accuracy achieved by the associative classifier Gamma, the average of the accuracy and Objects Retention Index, evaluated with other models from the state of the art such as the ENN, GGE, RNGE, MSEditB, all these methods based at classification error, which are considered of similar purpose. The evaluation was conducted on a set of international databases. Thus the results of the new method presented similar behavior with respect to the other models regarding the classification accuracy. By other hand the reduction of the number of instances obtained was superior in most of the databases.
Descripción
Palabras clave
Selección de Instancias, Clasificador Asociativo Gamma, Conjuntos Aproximados, Teoría Extendida, Inteligencia Artificial