Millimeter wave (mmWave) communication is being seriously considered for the next generation communication systems because of its ability to support high bandwidth and high data rates. Unfortunately, these systems perform badly in the presence of blockage. A sudden blockage in the line of sight(LOS) link leads to communication disconnection, which causes a reliability problem. Also, searching alternative base stations(BS) for re-connection results in latency overhead. In this paper, we tackle these problems by predicting the time of blockage occurrence using a machine learning (ML) technique. In our approach, BS learns how to predict that a certain link will experience blockage in the near future using the received signal power. Simulation results on a real dataset show that blockage occurrence can be predicted with 85% accuracy and the exact time instance of blockage occurrence can be obtained with low error. Thus the proposed method reduces the communication disconnections in mmWave communication, thereby increasing reliability and reducing latency of such systems.