In this paper, we propose the green stopping signal-centric predictive polling algorithm for machine type communications. Based on Markov optimal stopping theory, the green stopping signal-centric predictive polling algorithm exploits channel state information to reduce the energy consumption of wireless communications without knowing the mean channel gain. To reduce the computational complexity, the proposed algorithm is built upon a closed-form optimal stopping rule. Furthermore, we derive analytical results that characterize the optimal stopping time when the proposed green stopping signal-centric predictive polling algorithm is used. Simulation results show that the proposed approach could significantly reduce the average energy consumption at machines and the average signal prediction error at the base station in comparison with alternative approaches.