A semisupervised learning model based on fuzzy min–max neural networks for data classification

Abstract Semisupervised learning (SSL) models are useful for undertaking classification problems with a small set of labeled samples and a large number of unlabeled samples. In this regard, the family of fuzzy min–max (FMM) neural networks offers the capability of online learning for addressing both unsupervised and supervised problems. As such, this paper proposes a novel two-stage SSL model based on FMM networks, denoted SSL–FMM. The first stage employs the unlabeled samples to generate a number of hyperboxes using the unsupervised FMM algorithm, while the second stage uses the labeled samples to associate the generated hyperboxes with their target classes using the supervised FMM algorithm. In addition, a neighborhood-labeling mechanism based on the Euclidean distance and hyperbox centroids is formulated to associate the unlabeled hyperboxes with the most likely target classes. A number of benchmark problems and a real-world case study are employed to evaluate the effectiveness of the proposed SSL–FMM model. The outcome indicates that SSL–FMM is able to use unlabeled samples effectively and improve the FMM performance, producing promising results compared with other SSL methods in the literature.