Optimal Training Data Selection in Active Learning for Discrimination and Classification

Active learning has become a popular learning process for classification. By selecting the most beneficial training data, an active classifier achieves better classification accuracy than a passive classifier. We investigate the methods of developing two different types of optimal active learning processes, via either estimated discriminant functions or logistic regression. A comparison study is presented for the classifiers obtained by these methods. Performance of proposed active classifiers is evaluated under various conditions and assumptions. Optimal two-stage active learning is provided. Monte Carlo simulations have shown improved classification accuracy of our proposed active learning processes compared to passive learning process for all scenarios considered, with up to 10% accuracy improvement.