Evaluation of multi-label classifiers in various domains using decision tree

Abstract

One of the commonly used tasks in mining is classification, which can be performed using supervised learning approach. Because of digitization, lot of documents are available which need proper organization, termed as text categorization. But sometimes documents may reflect multiple semantic meanings, which represents multi-label learning. It is the method of associating a set of predefined classes to an unseen object depending on its properties. Different methods to do multi-label classification are divided into two groups, namely data transformation and algorithm adaptation. This paper focuses on the evaluation of eight algorithms of multi-label learning based on nine performance metrics using eight multi-label datasets, and evaluation is performed based on the results of experimentation. For all the multi-label classifiers used for experimentation, decision tree is used as a base classifier whenever required. Performance of different classifiers varies according to the size, label cardinality, and domain of the dataset.