Fusing absolute and relative information for augmenting the method of nearest neighbors for ordinal classification

Abstract

Listen

Translate article

Ordinal classification is a special case of multiclass classification in which there exists a natural order on the set of class labels. Due to the nature of the problem, datasets for ordinal classification are typically rather small, having a negative impact on performance. A possible way out is to look for additional information. In this paper, firstly, we make use of order relations for unlabeled examples to generate relative information. Secondly, we incorporate this relative information into the method of k nearest neighbors, thus exploiting absolute and relative information at the same time. More specifically, we bring together notions from the fields of information fusion and machine learning to integrate both types of information. Finally, we test the proposed method on some classical machine learning datasets. The experimental results show the effectiveness of our approach.