Multi-label classification using a fuzzy rough neighborhood consensus

A multi-label dataset consists of observations associated with one or more outcomes. The traditional classification task generalizes to the prediction of several class labels simultaneously. In this paper, we propose a new nearest neighbor based multi-label method. The nearest neighbor approach remains an intuitive and effective way to solve classification problems and popular multi-label classifiers adhering to this paradigm include the MLKNN and IBLR methods. To classify an instance, our proposal derives a consensus among the labelsets of the nearest neighbors based on fuzzy rough set theory. This mathematical framework captures data uncertainty and offers a way to extract a labelset from the dataset that summarizes the information contained in the labelsets of the neighbors. In our experimental study, we compare the performance of our method with five other nearest neighbor based multi-label classifiers using five evaluation metrics commonly used in multi-label classification. Based on the results on both synthetic and real-world datasets, we are able to conclude that our method is a strong competitor to nearest neighbor based multi-label classifiers like MLKNN and IBLR.