To Use the Relationships between Class Labels in Creating Classifier Chains to Improve Multi-label Classification

In this paper, we have supposed that there is meaningful relationships between the class labels in the multi-label classification problems and based on it, we have proposed two novel methods to improve the base classifier chains (CC) method for multilabel classification. In this paper, association rules are employed to determine the order of classifiers in the CC method for the first time. In the proposed methods, association rule mining is first employed to model the relationships between the class labels and then, an association graph is built based on the extracted rules and finally, the classifier chains is built based on the obtained graph. As there is meaningful relationships between the class labels in the real multi-label problems such as classifying the images and texts and medical applications, the proposed methods will improve the classification results in such contexts. Extensive experimental evaluations conducted on the benchmark datasets in the multi-label classification context show that to use the associations between the labels in constructing the classifier chains improves the results obtained by the main evaluation measures.