As the first theoretically proven capacity-achieving error correction code, polar codes have become a milestone in information theory field, which draw serious attention by its low complexity encoding and decoding structures. For the decoding section, studies are focused on efficient and low complex algorithms based on successive cancellation and belief-propagation (BP) decoders. To reduce computational complexity of BP-based decoders further, early stopping methods that avoid unnecessary iterations can be used. In this letter, we propose a reduced complexity early stopping method for BP-based polar code decoders. The proposed method is based on the hypothesis that observing only a small cluster of information bits polarized to the highest error probabilities is enough to detect successful decoding. Simulation results show that the proposed early stopping criterion significantly reduces the computational complexity of successful decoding detection as well as the total operation numbers of the whole decoding process compared with previous methods in the literature.