Human Stress Detection In and Through Sleep Using Machine Learning

Abstract

Human stress significantly affects sleep quality and overall health, making early detection essential for prevention and intervention. This study proposes a machine-learning-based approach to detect human stress during sleep using physiological and behavioural data such as heart rate, breathing patterns, body movements, and sleep stages. Data is collected through wearable devices and sleep-monitoring systems, then preprocessed and analysed using algorithms such as Support Vector Machines, Random Forest, and Deep Learning models. The system identifies patterns associated with stress, including irregular heart-rate variability and disturbed sleep cycles, and classifies stress into categories such as low, moderate, and high. Experimental results demonstrate that the proposed model can classify stress levels and provide real-time monitoring. This approach enables non-invasive, continuous stress assessment, helping individuals and healthcare providers take timely actions to improve mental well-being and sleep quality. The system follows a structured pipeline of data acquisition, preprocessing, feature extraction, model training, evaluation, and prediction, and was validated through nine functional test cases covering data input, preprocessing, feature extraction, training, evaluation, prediction, classification, and real-time monitoring, all of which passed. By leveraging physiological data and intelligent algorithms, the proposed system offers a reliable, efficient, and user-friendly approach to monitoring stress levels and supports early intervention and better quality of life.