Active learning method of bayesian networks classifier based on cost-sensitive sampling

Abstract

Bayesian networks classifier optimized by classification accuracy may have higher misclassification cost for imbalanced classification problem. Cost-sensitive learning method is aim to minimize classification cost. However, imbalanced training data consist of labeled and unlabeled samples in many classification tasks. So, active learning method based on cost-sensitive sampling is presented. Costsensitive loss function which is weighted with classification error loss function and classification cost loss function is proposed. Classification error loss function measures the classification accuracy of samples, and yet classification cost loss function measures the misclassification cost of samples. Then, active learning method of Bayesian networks classifier based on cost-sensitive sampling is proposed. Lastly, experiment results on a diagnostic dataset show that Bayesian networks classifier learned by active learning method based on cost-sensitive sampling is effectively in imbalanced dataset with labeled and unlabeled samples.