Active Clustering based Classification for Cost Effective Prediction in few Labeled Data Problem

In many data mining problems related to business, it is hard to obtain labeled instances. When the labeled data set is not large enough the classifiers often perform poor results. Nevertheless, semi-supervised learning algorithms, e.g. clustering based classification can learn from both labeled and unlabeled instances. The authors have planned and implemented a semi-supervised learning technique by combining the clustering based classification system with active learning. Their active clustering based classification method first clusters both the labeled and unlabeled data with the guidance of labeled instances, then queries the label of the most informative instances in an active learning cycle and after that classifies the data set. At cost benefit analysis comparing the results of their system with the supervised learning and clustering based classification it can be concluded that our solution saves the largest cost.