The present disclosure discloses a cerebral perfusion state classification apparatus, method and device, and a model training apparatus. In the apparatus, a transceiver module is used for receiving physiological feature data from different data collection devices; and a processor is used for extracting physiological features from the physiological feature data; inputting the physiological features into a random forest model to cause a plurality of decision-making trees in the random forest model to predict a cerebral perfusion state type corresponding to the physiological features; and classifying a cerebral perfusion state based on the cerebral perfusion state type corresponding to the physiological features.