Parallel and Sequential Support Vector Machines for Multi-label Classification

Multi-label classification is the problem that classes are not mutually exclusive, so that an example may belong to more than one category. This poses challenges to the traditional pattern recognition theory where class overlap means classification error. Multi-label classification arises typically in semantic scene classification, text categorization, medical diagnosis, and bioinformatics. However, only a small number of methods have been developed for multi-label classification. In this paper, we propose two algorithms, called Parallel Support Vector Machines (PSVMs) and Sequential Support Vector Machines (SSVMs), to handle multi-label classification problems. We applied them to scene classification. It is demonstrated that PSVM is comparable to, and SSVM outperforms the so-called cross-training C-criterion method. Keyword: multi-label classification; scene classification; parallel SVM; sequential SVM; crosstraining; c-criterion testing