Automatic labeling by means of semi-supervised fuzzy clustering as a boosting mechanism in the generation of fuzzy rules

Abstract

The ease of acquiring great volumes of data and the problems with manual labeling and interpretation of the output of learning results motivate the research on methods that can deal with the inherent aspects of this type of data. Studies suggest that semi-supervised learning is a possible alternative to obtain knowledge models that are more adequate to the reality of certain domains. Methods that make use of semi-supervision adapt traditional solutions to consider partial pre-existing information in the learning process. The goal of this work is to investigate a semi-supervised learning approach, involving semi-supervised fuzzy clustering as means to an automatic labeling mechanism. The studied approach uses the strategy previously proposed by the authors that includes a labeling process to increase the set of labeled examples used as input for a supervised fuzzy rule base generator. This paper presents and analyses additional results obtained from experiments designed to evaluate the impact of augmenting the labeled training examples on the final classification system.