Comparative Experimental Study of Multi Label Classification using Single Label Ground Truth with Application to Field Majoring Problem

Abstract

Researches on multi label classification methods generally use training data that already have multi label output as ground truth, but there are real-world problems where it is required to produce multi label prediction output but the available training data only have single label as ground truth. This study compared the performance of various multi label classification methods i.e. Ranking Support Vector Machine (Rank-SVM), Backpropagation for Multi Learning (BP-MLL), Multi Label K-Nearest Neighbor (ML-KNN), and Multi Label Radial Basis Function (ML-RBF) that were trained using multi label training data as intended and which were trained using single label training data. The dataset used in this research is an example of real-world problem, namely the personality-aptitude psychological test results is used to predict suitable majors in vocational high school. The results showed that hamming loss between the two was not far adrift so that it can be concluded that in certain problems, multi label classification methods can train single label and still produce multi label predictions with fairly good accuracy.