Hybridized Active Learning Model Based On Most Certain and Uncertain Label Selection

In order to correctly classify the huge amount of unlabeled data, supervised classification paradigms necessitated the requirement of labeled data. But the availability of labeled data is too scarce and labeling is too expensive. To decrease the human labeling efforts, through selecting the much meaningful data from unlabeled data and add to label data, active learning techniques have been proven to be efficient. Active Learning is based on the principle of selection of most uncertain and non-redundant instances in each iteration. In this paper, authors have considered not only the most uncertain instances but also the most certain instances have been selected which helped in improving the efficiency of learning model. Extensive experiment has been carried out on different datasets to confirm the effectiveness of proposed model.