Extensiones al ambiente de aprendizaje automatizado Weka-Parallel con distintos algoritmos de aprendizaje en redes bayesianas. Aplicaciones Bioinformáticas
Archivos
Fecha
2008-07-07
Autores
Arboláez Malboa, Asley
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Weka-Parallel es una versión de Weka, el cual es un software de aprendizaje
automatizado que permite realizar pruebas y validar algoritmos de la Inteligencia Artificial. Con el presente trabajo se adicionan algoritmos de búsqueda basados en Redes Bayesianas para clasificación usando la clase BayesNet de Weka, dos de ellos implementados anteriormente en otras versiones del Weka, siendo estos el PSO (Particle Swarm Optimization) y el BayesCHAID, además se implementa un tercer método de búsqueda, el BayNet. El primero de estos métodos utiliza la Optimización Basada en Enjambres de Partículas, el segundo usa el método estadístico Chi-cuadrado, y el tercero se basa de igual modo en este estadístico, pero usando árboles de decisión para construir las Redes Bayesianas. Se extiende el Weka-Parallel y no otra versión del software, debido a la facilidad que este nos
brinda de paralelizar los k-fold definidos durante la validación cruzada en cada uno de los experimentos. Se hace una comparación del nuevo algoritmo implementado con otros algoritmos para el aprendizaje de redes bayesianas, y por último se hacen experimentos con dos aplicaciones bioinformáticas para probar la eficiencia del Weka-Parallel usando los algoritmos adicionados al mismo.
Weka-Parallel is a version of Weka, which is a machine learning software developed to test and validate Artificial Intelligence’s algorithms. In this work some searching algorithms based on Bayesian networks for classification using BayesNet class of Weka are added. Two of them were implemented in other versions of this software, these are the PSO (Particle Swarm Optimization) and BayesCHAID. Now a third search method is implemented: BayNet. The first of these methods uses optimization based on swarms of particles, the second uses the statistical method Chi-square, and the third is similarly based in this statistical, but using decision trees to build the Bayesian Networks. This work extends Weka-Parallel, and not other version of this software, because of the facilities that it gives us in parallelization of the k-fold defined during the cross-validation. A comparison of the new algorithm implemented with other algorithms for learning Bayesian networks is made, and finally we show experiments with two bioinformatic applications to test the efficiency of Weka-Parallel using algorithms added to this program.
Weka-Parallel is a version of Weka, which is a machine learning software developed to test and validate Artificial Intelligence’s algorithms. In this work some searching algorithms based on Bayesian networks for classification using BayesNet class of Weka are added. Two of them were implemented in other versions of this software, these are the PSO (Particle Swarm Optimization) and BayesCHAID. Now a third search method is implemented: BayNet. The first of these methods uses optimization based on swarms of particles, the second uses the statistical method Chi-square, and the third is similarly based in this statistical, but using decision trees to build the Bayesian Networks. This work extends Weka-Parallel, and not other version of this software, because of the facilities that it gives us in parallelization of the k-fold defined during the cross-validation. A comparison of the new algorithm implemented with other algorithms for learning Bayesian networks is made, and finally we show experiments with two bioinformatic applications to test the efficiency of Weka-Parallel using algorithms added to this program.
Descripción
Palabras clave
Algoritmos de Aprendizaje, Aplicaciones Bioinformáticas