A Privacy Protected Sleep Quality Prediction Approach Using Ultrawideband Ambient Sensor,Synthetic Data,and Deep Learning

Sleep quality assessment is crucial for maintaining overall well-being and health. This paper proposes a novel approach for modeling sleep quality based on ultrawideband sensor data and machine learning techniques. The methodology involves data collection using an ultrawideband sensor placed near the subject's bed to record respiration patterns throughout the night. Due to limited available data, synthetic data generation techniques were employed to augment the dataset. The synthesized data was then categorized into specific sleep quality categories (Bad, Okay, Good) using the K-Nearest Neighbors (K-NN) algorithm based on Euclidean distance. Following data preprocessing and categorization, bidirectional Long Short-Term Memory (LSTM) models were utilized to model the relationship between respiration patterns and sleep quality. LSTM models are well-suited for sequential data processing and have shown promising results in various time-series prediction tasks. Hence, the performance of the proposed approach was evaluated through experiments using augmented real and synthetic data, and the results demonstrated the effectiveness of the bidirectional LSTM models with reasonable accuracy. Thus, utilizing ultrawideband sensor-based respirations and synthetic data combined with machine learning techniques can contribute to objective sleep quality assessment.