Experimental Comparison of Different Problem Transformation Methods for Multi-Label Classification using MEKA

Abstract

Classification of multi-label and multi-target data is challenging task for machine learning community. It includes converting the problem in other easily solvable form or extending the existing algorithms to directly cope up with multi-label or multi-target data. There are several approaches in both these category. Since this problem has many applications in image classification, document classification, bio data classification etc. much research is going on in this specific domain. In this paper some experiments are performed on real multi-label datasets and three measures like hamming loss, exact match and accuracy are compared of different problem transformation methods. Finally what is effect of these results on further research is also highlighted.