Multi-label classification models for heterogeneous data: an ensemble-based approach.

In recent years, the multi-label classification gained attention of the scientific community given its ability to solve real-world problems where each instance of the dataset may be associated with several class labels simultaneously, such as multimedia categorization or medical problems. The first objective of this dissertation is to perform a thorough review of the state-of-the-art ensembles of multi-label classifiers (EMLCs). Its aim is twofold: 1) study state-of-the-art ensembles of multi-label classifiers and categorize them proposing a novel taxonomy; and 2) perform an experimental study to give some tips and guidelines to select the method that perform the best according to the characteristics of a given problem. Since most of the EMLCs are based on creating diverse members by randomly selecting instances, input features, or labels, our main objective is to propose novel ensemble methods while considering the characteristics of the data. In this thesis, we propose two evolutionary algorithms to build EMLCs. The first proposal encodes an entire EMLC in each individual, where each member is focused on a small subset of the labels. On the other hand, the second algorithm encodes separate members in each individual, then combining the individuals of the population to build the ensemble. Finally, both methods are demonstrated to be more consistent and perform significantly better than state-of-the-art methods in multi-label classification