Accurately predicting and inferring a driver's decision to brake is criticalfor designing warning systems and avoiding collisions. In this paper we focuson predicting a driver's intent to brake in car-following scenarios from aperception-decision-action perspective according to his/her driving history. Alearning-based inference method, using onboard data from CAN-Bus, radar andcameras as explanatory variables, is introduced to infer drivers' brakingdecisions by combining a Gaussian mixture model (GMM) with a hidden Markovmodel (HMM). The GMM is used to model stochastic relationships among variables,while the HMM is applied to infer drivers' braking actions based on the GMM.Real-case driving data from 49 drivers (more than three years' driving data perdriver on average) have been collected from the University of Michigan SafetyPilot Model Deployment database. We compare the GMM-HMM method to a supportvector machine (SVM) method and an SVM-Bayesian filtering method. Theexperimental results are evaluated by employing three performance metrics:accuracy, sensitivity, specificity. The comparison results show that theGMM-HMM obtains the best performance, with an accuracy of 90%, sensitivity of84%, and specificity of 97%. Thus, we believe that this method has greatpotential for real-world active safety systems.