Abstract
[Display omitted]
An example search space of hmAntMiner-C for constructing rule antecedent.
•This paper presents a hierarchical multi-label classification algorithm (hmAntMiner-C).•It uses correlation of attribute-value pairs for constructing IF-THEN rule list.•Comparison is provided with some other state of the art algorithms with promising results.
There exist numerous state of the art classification algorithms that are designed to handle the data with nominal or binary class labels, where a sample belongs to only a single class label. In these problems, known as flat classification problems, class labels are independent of each other. Unfortunately, on the other hand, less attention is given to the genre of classification problems where samples may belong to several classes and at the same time the class labels are organized based on a structured hierarchy; such as gene ontology, protein function prediction, test scores, web page categorization, text categorization etc. This article presents a novel Ant Colony Optimization based hierarchical multi-label classification algorithm that can handle such a complex instance of classification problems and can incorporates the given class hierarchy during its learning phase. The algorithm produces IF-THEN ordered rule list to learn a comprehensible model which can easily be verified by experts. It exploits positive correlation between the domain values of two related attributes to improve the discrimination power of resultant classification model, up to a significant level. The paper contains rich details regarding hierarchical single label (or single path) and multi-label classification problems and different categories of corresponding solutions. The proposed method is evaluated on sixteen most challenging bioinformatics datasets; some of these containing hundreds of attributes and thousands of class labels. At the end, the proposed method is compared with four recent state of the art hierarchical multi-label classification algorithms. The empirical evaluation confirms the promising ability of the proposed technique for hierarchical multi-label classification task.