A Hybrid Classification Technique using Belief Rule Based Semi-Supervised Learning

An advancement in the paradigm of machine learning has been acclaimed by the arrival of semi-supervised learning. In real life, it is challenging to get enough labeled samples. On the other hand, unlabeled data are inexpensive because it does not need human expertise. As a result, a lot of research attention has been gained in the semi-supervised field as both labeled and unlabeled data are greatly utilized. Traditional semi-supervised learning may experience information loss and problems with uncertainty. Moreover, the samples cause ambiguity because of unclear descriptions of the information and vague evidence. Therefore, misclassification occurs. In this paper, the ambiguity is dealt by merging the belief rule based methodology with semi-supervised self-training. To further study, the datasets were categorized into three ratios which show that our proposed method works significantly better than the traditional approach. Moreover, this research has included a comparison of results between the proposed and the traditional approach.