Active learning is a kind of semi-supervised learning methods in which learning algorithm is able to interactively query some information to get new subjects’ labels/classes. When labeling subjects is quite expensive, active learning is a possible solution to reduce cost because only the selected subjects need to be exanimated and labeled, such as in money laundering detection and disease screening. For analyzing large-scale datasets, the large sample size and high dimension become a challenge for both analysis and computation. In this talk, we will present an active learning algorithm for analyzing large-scale datasets. The proposed method is based on a logistic regression model with a modified iterative algorithm for estimating parameters in order to be more computational efficiency, without sacrificing too much in statistical efficiency. In addition, the methods of shrinkage estimation and subject clustering are considered for selecting effective variables and reducing subject-searching time when analyzing large-scale datasets. For the perspectives of uncertainty sampling and precision of parameter estimates, we search the representatives of subject clusters and select useful samples based on the concept of sequential D-optimal design. The real data applications and simulations will be used to evaluate the performance of the proposed active learning algorithm.