Abstract: A semi-supervised active learning algorithm is proposed to overcome the difficulties in getting sufficient labeled training sample data set during the process of building dynamic Bayesian network(DBN) classification model.Although semi-supervised learning can use unlabeled sample data to learn DBN classification model,it often suffers from adding incorrect class information which affects the accuracy of classification model in the iteration process.Active learning combined with semi-supervised learning can maximally increase the accuracy of classifying unlabeled samples through actively selecting useful unlabeled samples for the user to label and adding them to the training data.Experiment results show that the proposed algorithm can notably improve the efficiency and accuracy of the DBN learner and can enable the model to quickly achieve the expected classification precision.