This paper analyzes the differences between obstructive sleep apnea syndrome (OSAS) patients and healthy subjects focusing on the WAKE stage (shallow sleep) classification rule generated by Random Forests. In particular, our method generates the classification rules of WAKE/Non-WAKE stage for each subject from the power spectrums of mattress sensor outputs and extracts the most used frequency for the classification of WAKE/Non-WAKE stage. In cooperation with medical institutions, we conducted the experiments with nine OSAS subjects (four mild patients and five moderate patients) and nine healthy subjects (age 20 to 60) to analyze the differences between OSAS subjects and healthy subjects. The analysis of the rules derives the following implications: (1) the high frequencies (more than 1Hz) which are affected by the large body movements tend to be employed for the classification of WAKE/Non-WAKE in healthy subjects; (2) the low frequencies (i.e., 0.90 Hz, 0.53 Hz and 0.31 Hz) which are included in the range of respiration and heart beats but not directly related to them tend to be employed for the classification in OSAS subjects.