Fuzzy support vector machines for multilabel classification

• We develop a fuzzy support vector machine (FSVM) for multilabel classification.
• We resolve unclassifiable regions by membership functions.
• We also resolve undefined multilabels by membership functions.
• Effectiveness of the proposed FSVM is demonstrated by computer experiments.

The problem of one-against-all support vector machines (SVMs) for multilabel classification is that a data sample may be classified into a multilabel class that is not defined or it may not be classified into any class. To solve this problem, in this paper we propose fuzzy SVMs (FSVMs) for multilabel classification, in which for each multilabel class, a region with the associated membership function is defined and a data point is classified into a multilabel class whose membership function is the largest. By computer experiments, we show that the accuracy is improved by the FSVM over the conventional one-against-all SVM.