Multi-label classification with Bayes' theorem
Guangzhi Qu, Hui Zhang, Craig T. Hartrick · 2011
Compared with single-label classification, multi-label classification is more general in practice, since it allows one instance to have more than one label simultaneously. Bayes' Theorem has been successfully applied to deal with single-label classification. In this paper, we proposed to tackle multi-label classification using Bayes' Theorem. We propose two approaches, coined as Pair-Dependency Multi-Label Bayesian Classifier (PDMLBC) and Complete-Dependency Multi-Label Bayesian Classifier (CDMLBC). PDMLBC takes advantage of label dependency between any two labels, while CDMLBC considers the dependency among a set of labels. In the experiments, we evaluate the performance of PDMLBC and CDMLBC on real medical data, the results show that both PDMLBC and CDMLBC methods outperform NB+BR on all metrics, and CDMLBC works best among the three methods.