Explainable Non-Contact Sleep Apnea Syndrome Detection Based on Comparison of Random Forests

Abstract

This paper focuses on Sleep Apnea Syndrome (SAS) and proposes the novel eXplainable AI (XAI) method that extracts characteristics of SAS by comparing the datasets of the SAS patients and the non-SAS subjects. For this issue, this paper (i) employs “two” Random Forests (RFs) to respectively learn the models for the SAS patients and the non-SAS subjects to classify the WAKE/non-WAKE stage, (ii) compares the two learned RFs to find their difference as the physiological characteristic of SAS, and (iii) proposes the SAS detection method based on the difference between the two learned RFs. Through the human subject experiment of the SAS detection based on the biological vibration data acquired from the mattress sensor during sleep, the following implications have been revealed: 1) RF learned from the SAS patient data classifies the WAKE/non-WAKE stage from the viewpoint of the “low” frequencies of the biological vibration data, while RF learned from the non-SAS subject data classifies it from the viewpoint of its “high” frequencies; and 2) the SAS patients have the WAKE stage with the low frequencies of the biological vibration data caused by disturbances in the autonomic nervous system due to apnea/hypopnea, while the non-SAS subjects do not have it but have the usual WAKE stage with the high frequencies caused by large body movements, which has a potential of the new characteristic of SAS instead of respiration as a traditional characteristic of SAS.

Keywords

• XAI
• random forests
• sleep apnea syndrome
• feature importance
• mattress sensor