The main area considered in this package is the semi-supervised learning (SSL). SSL is a suitable approach when we dispose of a limited number of labeled examples and a great amount of unlabeled data (Chapelle, 2006). Specifically, semi-supervised classification (SSC) focuses on training a classifier such that it outperforms a supervised classifier trained on the labeled data alone. In semi-supervised classification, the dataset can be divided into two parts, L and U. Let L be the set of instances Xl=(x1,...,xl) for which the labels Yl=(y1,...,yl) are provided. Let U be the set of instances Xu=(xl+1,...,xl+u) for which the labels are not known. We follow the typical assumption that there is much more unlabeled than labeled data, i.e., u >> l. The whole set L union U forms the training set.
An specific family of SSC methods, denoted self-labeled techniques (Triguero, 2015), aims to enlarge the original labeled set using the most confident predictions to classify unlabeled data. In contrast to other approaches, self-labeled techniques do not make any special assumptions about the distribution of the input data. All methods that have been implemented in this R-package belong to this family of SSC methods.