Publication Date


Document Type


Committee Members

Tanvi Banerjee, Ph.D. (Advisor); Mateen M. Rizki, Ph.D. (Committee Member); Krishnaprasad Thirunarayan, Ph.D. (Committee Member); William Romine, Ph.D. (Committee Member); Ali Azarbarzin, Ph.D. (Committee Member)

Degree Name

Doctor of Philosophy (PhD)


Humans spend almost a third of their lives asleep. Sleep has a pivotal effect on job performance, memory, fatigue recovery, and both mental and physical health. Sleep quality (SQ) is a subjective experience and reported via patients’ self-reports. Predicting subjective SQ based on objective measurements can enhance diagnosis and treatment of SQ defects, especially in older adults who are subject to poor SQ. In this dissertation, we assessed enhancement of subjective SQ prediction using an easy-to-use E4 wearable device, machine learning techniques and identifying disease-specific risk factors of abnormal SQ in older adults. First, we designed a clinical decision support system to estimate SQ and feeling refreshed after sleep using data extracted from an E4 wearable device. Specifically, we processed four raw physiological signals of heart rate variability (HRV), electrodermal activity, body movement, and skin temperature using distinct signal processing methodologies. Following this, we extracted signal-specific features and selected a subset of the features using recursive feature elimination cross validation strategy to maximize the accuracy of SQ classifiers in predicting the SQ of older caregivers. Second, we investigated discovering more effective features in SQ prediction using HRV features which are not only effortlessly measurable but also can reflect sleep stage transitions and some sleep disorders. Evaluation of two interpretable machine learning methodologies and a convolutional neural network (CNN) methodology demonstrated the CNN outperforms by an accuracy of 0.6 in predicting light, medium, and deep SQ. This outcome verified the capability of using HRV features measurable by easy-to-use wearable devices, in predicting SQ. Finally, we scrutinized daytime sleepiness risk factors as a sign of abnormal SQ from four perspectives: sleep fragmented, sleep propensity, sleep resilience, and non-restorative sleep. The analysis demonstrates distinguishability of the main risk factors of excessive daytime sleepiness (EDS) between patients suffering from fragmented sleep (e.g. apnea) and sleep propensity. We identified the average area under oxygen desaturation curve corresponds to apnea/hypopnea event as a disease-specific risk factor of abnormal SQ. Our further daytime sleepiness prediction demonstrated the significant role of the founded disease-specific risk factor as well.

Page Count


Department or Program

Department of Computer Science and Engineering

Year Degree Awarded


Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.