Scholarlink

Purpose <p>Restful sleep depends on physical, physiological, and behavioral factors. Traditional methods, such as polysomnography, monitor brain waves, eye movements, muscle activity, heart rhythm, and breathing patterns; however, these methods are expensive, complex, and limited to clinical settings. An alternative approach, the Pittsburgh Sleep Quality Index (PSQI), is a subjective tool in which the score is derived from a questionnaire and may be prone to errors. To address the limitations of existing methods, the proposed approach extracts features from physical and physiological data, analyzes the importance of these features to assess sleep quality, and classifies the data into good and poor sleep quality categories.</p> Methods <p>Data for the present study were obtained from the multilevel monitoring of activity and sleep in healthy individuals dataset. Sleep quality was categorized as good or poor using features such as heart rate, RR interval, and three-axis movement data. Machine learning models were applied, and their performance was evaluated using five-fold cross-validation and subsequently tested on unseen data. Furthermore, feature importance was determined using Shapley additive explanations (SHAP) values. Finally, the best-performing classifier was cross-validated using only the most important features.</p> Results <p>The gradient boosting machine (GB) and decision tree (DT) models outperformed other classifiers in the five-fold cross-validation, achieving mean validation accuracies of 98.98 and 90.74%, respectively. On unseen data, both models achieved 100% testing accuracy, with mean prediction times per sample of 10.7&#xa0;µs for GB and 3.78&#xa0;µs for DT. Notably, using only the top three features yielded an accuracy greater than 97%, while using the top nine and top eight features resulted in accuracies of 99.5 and 99.9%, respectively.</p> Conclusion <p>This study demonstrates that objective analysis of sleep quality using physical and physiological features can achieve high accuracy and may be considered for future clinical validation in patients with sleep disorders.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Physical and physiological data analysis for sleep quality assessment and performance evaluation of machine learning approaches

  • 期刊论文

  • 发表日期：

    2025年9月23日

查看全文
  • Srinivasulu Avvaru,
  • Sushma TV,
  • Susmitha Alamuru,
  • Ramesha M,
  • Anjali G,
  • Srinivasulu Boyineni

摘要

Purpose

Restful sleep depends on physical, physiological, and behavioral factors. Traditional methods, such as polysomnography, monitor brain waves, eye movements, muscle activity, heart rhythm, and breathing patterns; however, these methods are expensive, complex, and limited to clinical settings. An alternative approach, the Pittsburgh Sleep Quality Index (PSQI), is a subjective tool in which the score is derived from a questionnaire and may be prone to errors. To address the limitations of existing methods, the proposed approach extracts features from physical and physiological data, analyzes the importance of these features to assess sleep quality, and classifies the data into good and poor sleep quality categories.

Methods

Data for the present study were obtained from the multilevel monitoring of activity and sleep in healthy individuals dataset. Sleep quality was categorized as good or poor using features such as heart rate, RR interval, and three-axis movement data. Machine learning models were applied, and their performance was evaluated using five-fold cross-validation and subsequently tested on unseen data. Furthermore, feature importance was determined using Shapley additive explanations (SHAP) values. Finally, the best-performing classifier was cross-validated using only the most important features.

Results

The gradient boosting machine (GB) and decision tree (DT) models outperformed other classifiers in the five-fold cross-validation, achieving mean validation accuracies of 98.98 and 90.74%, respectively. On unseen data, both models achieved 100% testing accuracy, with mean prediction times per sample of 10.7 µs for GB and 3.78 µs for DT. Notably, using only the top three features yielded an accuracy greater than 97%, while using the top nine and top eight features resulted in accuracies of 99.5 and 99.9%, respectively.

Conclusion

This study demonstrates that objective analysis of sleep quality using physical and physiological features can achieve high accuracy and may be considered for future clinical validation in patients with sleep disorders.