Publication Date

Fall 2025

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Sayma Akther

Second Advisor

Navrati Saxena

Third Advisor

Sai Vamsi Dutt Patibandla

Keywords

Sleep Stage Prediction, Machine Learning, Time-Series Decomposition, TimeGAN, Ensemble Models

Abstract

The process of determining health-related sleep stages through laboratory independent methods has not been developed at a scalable level. The gold-standard labels from polysomnography (PSG) require laboratory-based monitoring which restricts its use in real-world applications. The current wearable-based sleep classification systems detect only binary sleep/wake states while ignoring the complete range of daytime activities. The proposed method solves existing problems through three main components which include 24-hour wrist actigraphy decomposition into statistical features across different time periods and TimeGAN (TGAN) based synthetic time series generation for sleep stage imbalance correction and reinforcement learning-based feature selection and optimization. The evaluation process uses eight MESA participants who received actigraphy and PSG label pairs to demonstrate that the method produces better than 0.75 accuracy and better than 0.70 macro-averaged F1 scores while maintaining high minority-stage detection rates. The research shows that it is possible to track detailed sleep patterns through daily movement tracking which will enable developers to create home-based sleep monitoring systems. The research team plans to increase participant numbers and add multiple sensors and develop advanced deep learning models for better results.

Download

Available for download on Saturday, December 19, 2026

Share

COinS