Department of Electrical and Computer Engineering, The University of Memphis, Memphis, TN 38152, USA.
Sensors (Basel). 2023 May 31;23(11):5220. doi: 10.3390/s23115220.
Approximately 65% of the worldwide adult population has experienced stress, affecting their daily routine at least once in the past year. Stress becomes harmful when it occurs for too long or is continuous (i.e., chronic), interfering with our performance, attention, and concentration. Chronic high stress contributes to major health issues such as heart disease, high blood pressure, diabetes, depression, and anxiety. Several researchers have focused on detecting stress through combining many features with machine/deep learning models. Despite these efforts, our community has not agreed on the number of features to identify stress conditions using wearable devices. In addition, most of the reported studies have been focused on person-specific training and testing. Thanks to our community's broad acceptance of wearable wristband devices, this work investigates a global stress detection model combining eight HRV features with a random forest (RF) algorithm. Whereas the model's performance is evaluated for each individual, the training of the RF model contains instances of all subjects (i.e., global training). We have validated the proposed global stress model using two open-access databases (the WESAD and SWELL databases) and their combination. The eight HRV features with the highest classifying power are selected using the minimum redundancy maximum relevance (mRMR) method, reducing the training time of the global stress platform. The proposed global stress monitoring model identifies person-specific stress events with an accuracy higher than 99% after a global training framework. Future work should be focused on testing this global stress monitoring framework in real-world applications.
大约 65%的全球成年人口经历过压力，至少在过去一年中，有一次影响了他们的日常生活。当压力持续时间过长或持续存在（即慢性）时，就会变得有害，干扰我们的表现、注意力和专注力。慢性高压力会导致严重的健康问题，如心脏病、高血压、糖尿病、抑郁和焦虑。一些研究人员专注于通过将许多特征与机器/深度学习模型相结合来检测压力。尽管做出了这些努力，但我们的社区尚未就使用可穿戴设备识别压力状况所需的特征数量达成一致。此外，大多数报告的研究都侧重于特定于个人的培训和测试。由于我们的社区广泛接受可穿戴腕带设备，这项工作调查了一个结合了八个 HRV 特征和随机森林 (RF) 算法的全球压力检测模型。虽然模型的性能针对每个人进行评估，但 RF 模型的训练包含所有受试者的实例（即全球训练）。我们使用两个开放访问数据库（WESAD 和 SWELL 数据库）及其组合验证了所提出的全球压力模型。使用最小冗余最大相关性 (mRMR) 方法选择具有最高分类能力的八个 HRV 特征，从而减少了全球压力平台的训练时间。所提出的全球压力监测模型在经过全球训练框架后，能够以高于 99%的准确率识别特定于个人的压力事件。未来的工作应集中在真实应用中测试这个全球压力监测框架。
