The classification learner app will not use thresholds between the fatigue levels in that system. Some human went through all of the signals and classified sections by hand, possibly based on information that is not available in the data you are provided, such as video monitoring or discussion with the driver. The data has been labeled somehow and now it is the task of the learner app to find something different in the signals between the states.
The combinations that are noticed for each of the states do not necessarily have anything odd about the individual signals. For example it might be the case that the P wave is in a certain range if you are wide awake and happy, but that if the "happy" is absent then that same P value might indicate high fatigue. Thresholds do not necessarily apply