Using Gaussian Naïve Bayes in Active Learning

Nowadays supervised approach has many applications in different classification problems. Although these approaches have achieved many successes, they need large amount of training data. In this type of approach, all of the samples in dataset should be labeled which will be costly. Moreover in most of supervised learning approaches samples will be chosen randomly and because training is based on samples with similar data provided to system before will have less training effect on model. So a large amount of training data have low training effect on training process. In addition if training is based on a special type of frequently encountered samples, overfitting problem will occur. Active learning is presented to solve this problem in which samples with the most favorable impact on training process are selected repeatedly. Active learning tries to achieve maximum learning effect by utilizing a goal-driven approach with minimum number of training samples and subsequently minimum amount of labeling. Correct utilization of Active Learning can turn it into a powerful tool to cope with situation in which we are faced with deficient labeled data. In this paper, we propose a new model using active learning and Gaussian Naïve Bayes algorithm, which increases the accuracy of classification with a small number of training samples and the problem of overfitting does not occur