Publication: Predicción y selección de características, mediante análisis local de la fiabilidad, para el mercado de valores y su extensión a problemas de clasificación y regresión
Loading...
Identifiers
Publication date
2017-05
Defense date
2017-09-08
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Esta tesis se encuadra dentro del ámbito del Aprendizaje Automático, un
área de la Inteligencia Artificial (IA). A lo largo de la misma, se han diseñado
y validado experimentalmente, nuevas técnicas de selección de atributos y
de clasificación.
La motivación para el desarrollo de dichas técnicas, se basa en el deseo de
implementar herramientas adecuadas para tratar problemas de selección de
atributos y de clasificación en un dominio de especial dificultad: el mercado
de valores.
Se ha partido de la hipótesis de que los factores que dificultan la clasificación
correcta de los datos son, a menudo, una ratio desfavorable entre información
y ruido, una alta dimensionalidad, escasez de patrones y desbalanceo
del número de patrones de cada clase. Una vez identificados dichos factores,
se han diseñado técnicas robustas frente a estos, concretamente un algoritmo
de selección de atributos (con diferentes variantes) y un algoritmo de
clasificación.
Estas técnicas se han validado sobre un exhaustivo conjunto de problemas
generados artificialmente y en problemas reales del mercado de valores.
Por último, se ha explorado la posibilidad de utilizar las nuevas técnicas de
selección de atributos propuestas en problemas convencionales. Para ello,
se han validado sobre un conjunto de dominios reales de uso común en
Aprendizaje Automático, tanto para clasificación como para regresión.
This thesis belongs to Machine Learning, an area of Artificial Intelligence (AI). During its development, new techniques of attribute selection and classification have been designed and validated empirically. The motivation for the development of these techniques is based on the desire to implement adequate tools to deal with feature selection and classification problems in an area of particular difficulty: the stock market. Based on the hypothesis that the factors which make data classification difficult are, frequently, a low ratio between information and noise; high dimensionality, small training samples, and class imbalance. Once these factors have been identified, robust techniques to deal with them were designed, specifically a feature selection algorithm (with different variants) and a classification algorithm. These techniques have been validated over exhaustive synthetic data sets and stock market problems. Finally, the possibility of using the new feature selection techniques were explored in conventional problems. To this end, they were validated using a data set of actual domains, both for classification and regression.
This thesis belongs to Machine Learning, an area of Artificial Intelligence (AI). During its development, new techniques of attribute selection and classification have been designed and validated empirically. The motivation for the development of these techniques is based on the desire to implement adequate tools to deal with feature selection and classification problems in an area of particular difficulty: the stock market. Based on the hypothesis that the factors which make data classification difficult are, frequently, a low ratio between information and noise; high dimensionality, small training samples, and class imbalance. Once these factors have been identified, robust techniques to deal with them were designed, specifically a feature selection algorithm (with different variants) and a classification algorithm. These techniques have been validated over exhaustive synthetic data sets and stock market problems. Finally, the possibility of using the new feature selection techniques were explored in conventional problems. To this end, they were validated using a data set of actual domains, both for classification and regression.
Description
Keywords
Aprendizaje automático, Inteligencia artificial, Selección de atributos, Clasificación, Predicción, Mercado de valores