Selective Sampling for Classification

One of the goals of machine learning researches is to build accurate classifiers form an amount of labeled examples. In some problems, it is necessary to gather a large set of labeled examples which can be costly and time-consuming. To reduce these expenses, one can use active learning algorithms. These algorithms benefit from the possibility of performing a small number of label-queries from a large set of unlabeled examples to build an accurate classifier. It should be mentioned that actual active learning algorithms, themselves, have some known weak points which may lead them to perform unsuccessfully in certain situations. In this thesis, we propose a novel active learning algorithm. Our proposed algorithm not only fades the weak points of the previous active learning algorithms, but also performs competitively among the widely known active learning algorithms while it is easy to implement.