Redes de arquitectura profunda y ensembles para el tratamiento de la alta dimensionalidad y el desbalanceo en aprendizaje supervisado

Abstract

This thesis deals with the study of a new and promising field, the use of new techniques based on deep learning and ensembles that address the problems of high dimensionality and imbalance of data. This choice is due to the significant raise they have experienced in recent years, offering relevant results in many fields of application. Furthermore, the reason for facing the two problems mentioned above is that the characteristics inherent in the data are constantly changing and the tendency is that both dimensionality and imbalance continue to increase. The work focuses on addressing the task of dimensionality reduction through the use of Autoencoder (AE). In this sense, both experimental analysis of existing models and new proposals for classification methods based on AE are carried out. However, this thesis also faces the problem of imbalance from the perspective of deep learning.