Active learning for cost-sensitive classification using logistic regression model
Siyuan Zhou, Ya Zhang · 2016
Active learning aims to selectively label the most informative examples to save the data collection cost. While active learning has been well studied for balanced classification problems, limited research is performed in cost-sensitive scenario. In this paper, we investigate the problem of active learning for cost-sensitive classification. We first propose a general active learning framework named GEM, which chooses examples leading to the minimum generalization error. Then we incorporate the misclassification cost into expected loss calculation under the proposed framework, and derive a model estimation rule with the Newton-Raphson method using logistic regression as the base model. Finally, we present the complete active learning algorithm for cost-sensitive classification. Extensive experiments on various benchmark data sets from the UCI repository have demonstrated the effectiveness of the proposed algorithm.