Improving Sleep Quality Classification Through Synthetic Sensor-Based Medical Data Generation

Sleep quality is a critical factor that influences physical, mental, and emotional health. Advances in medical devices and machine learning offer promising opportunities to improve sleep monitoring systems. However, limited data availability and strongly underrepresented classes hinder the development of reliable models. Synthetic data generation has emerged as a potential solution by increasing the size and heterogeneity of the training set. This study presents a system to assess sleep quality using cardiac data collected by wearable sensors and explores the effectiveness of various synthetic data generation methods. The methods were evaluated based on generation time, statistical similarity to real data, and their impact on the machine learning model’s performance. The experimental results show that synthetic data generation methods, particularly the CTGAN, significantly improve sleep quality classification performance, especially for minority classes.