The present paper proposes a seismic damage prediction method for high-speed railway bridges (HSRBs) based on active learning (AL) technology, which can achieve good prediction performance by selecting the most informative samples for labeling and using these samples to train the machine learning (ML) model. In this process, an extreme gradient boosting (XGBoost) model is used as the learner, and three query strategies are considered for AL. Then, the AL-XGBoost model is developed and its prediction performance is further evaluated. The results show that the AL-XGBoost model using batch mode sampling strategy through training with fewer labeled samples has comparable the performance of seismic damage prediction to the XGBoost model developed with complete labeled samples.