A reliable ensemble based approach to semi-supervised learning

Semi-supervised learning (SSL) methods attempt to achieve better classification of unseen data through the use unlabeled than can be achieved by from available labeled alone. Most SSL require user familiarize themselves with novel, complex concepts and ensure underlying assumptions made these match problem structure, or they risk a decrease in predictive performance. In this paper, we present reliable semi-supervised ensemble (RESSEL) method, which exploits using it generate diverse classifiers self-training combines into an for prediction. Our method functions as wrapper around supervised base classifier refrains introducing additional dependent assumptions. We conduct experiments on number commonly used sets prove its merit. The results show RESSEL improves significantly upon alternatives, provided that is able produce adequate probability-based rankings. It shown delivers comparable if requirement not met, while also broadens range good parameter values. Furthermore, demonstrated outperform existing self-labeled approaches.