Machine Learning Methods for Ordinal Classification with Additional Relative Information

Mengzi Tang, Raúl Pérez-Fernández, Bernard De Baets

Abstract: The ability to learn an accurate classification process is often limited by the amount of labeled data. Incorporating additional information into the learning process for overcoming this limitation has been a popular research topic. In this work, we focus on ordinal classification problems that are provided with limited absolute information and additional relative information. We modify some classical machine learning methods to combine both types of information. Moreover, we propose a new distance metric learning method to exploit both types of information for learning a suitable distance metric that can be incorporated into the augmented method of k nearest neighbors for ordinal classification. The experimental results show that our method is competitive with other modified machine learning methods and considering additional relative information leads to a better performance.