Stress measurement using physiological signals has been performed in the literature using electrocardiography (ECG) [25], electroencephalography (EEG) [26], photoplethysmography (PPG) [27], galvanic skin response (GSR) [28], skin temperature (ST) [29], heart rate (HR) [30], heart rate variability (HRV) [31], blood volume pressure (BVP) [32], electromyogrpahy [33] and by fusion of these bio-signals [34]. In literature, several acute human stress measurement studies using different stimuli have been proposed. The framework used for the assessment of human stress in the literature follows the following steps. Firstly, physiological signals are acquired by inducing stress either in laboratory settings [35-38] or in real life situations [39-43]. Followed by this, a discriminating set of features are extracted to classify human stress by either using some pattern recognition or machine learning techniques. Subsequently, a diverse range of machine learning techniques has been used to classify human stress into two and three levels. The public speaking activity has been used as a stress-inducing stimulus in various stress measurement studies [44,45]. An acute stress measurement scheme using unobtrusive physiological markers of skin temperature, heart rate, and pulse wave in response to the mental arithmetic task and public speaking stimuli was presented in Ref. [11]. The study concluded that the rest state (27.86 auxiliary units) can be discriminated from stress condition (47.55 auxiliary units) using these unobtrusive physiological bio-markers. Another human stress measurement study using HRV signals was presented in Ref. [46] and data was annotated using scores of the STAI questionnaire. A human stress assessment scheme using GSR and PPG signals in response to stimuli of cognitive task and public speaking was presented in Ref. [47]. Human stress assessment using physiological signals (GSR, PPG, and HRV) and sociometric sensors (microphone, accelerometer, proximity, and infrared sensors) in response to a cognitive task and public speaking stimuli were presented in Ref. [48]. Another study for analyzing the real-life public speaking anxiety using recurrence quantification analysis of the GSR signals was presented in Ref. [49]. Non-linear features of the GSR signals and neural network were used to identify the calm and the anxious state as well as low anxiety and high anxiety state. Temporal parametric models to quantify the bio-behavioral trajectories of public speaking anxiety using real-life public speaking stimuli and physiological signals of GSR, PPG, and HR was presented in Ref. [50]. The study concluded that the proposed time-based analysis of the bio-behavioral measures was more robust in estimating the characteristics of the participants as compared to the traditional model. The multimodal human stress assessment methods have remarkable results in terms of obtained accuracy as compared to single modality based methods [51,52]. Multimodal stress assessment schemes in response to different stimuli have been proposed in a wide range of studies available in the literature. Human stress measurement using a fusion of near-infrared spectroscopy (NIRS) and physiological signals of PPG, ECG, and seismocardiogram in response to mental arithmetic and memory tasks was presented in Ref. [53]. An accuracy of 85% was achieved for three states using a random forest (RF) classifier. Assessment of mental stress using a fusion of EEG and functional-NIRS signals in response to mental arithmetic task with an accuracy of 95.1% using SVM classifier was proposed in Ref. [54]. The fusion of EEG signals with other psychological modalities like ECG [55] and galvanic skin response [56] has been used for boosting the performance of stress classification. The fusion of GSR and ST has been used to classify human stress with an accuracy of 84% [57]. Heart rate, skin conductance, and accelerometer sensors have been adopted for the measurement of real-time stress recognition using daily-life activities [58]. An accuracy of 92.15% was achieved for three-level stress classification using the fusion of features from these three modalities. A novel multimodal stress recognition scheme using a fusion of features from BVP, HR, ST, GSR, and respiratory rate (RR) in response to the mental arithmetic task has been proposed in Ref. [59]. An accuracy of 83% and 72% is achieved for two and three class stress classification, respectively. Some of the other stress measurement studies available in the literature include fusion of ECG and contextual data obtained from smartphones [60], a fusion of accelerometer and electrodermal activity (EDA) data [61], a fusion of respiratory, ECG and accelerometer sensors data [62] and fusion of accelerometer, EDA and Bluetooth sensors data [63].🏁

Stress measurement using physiological signals has been performed in the literature using electrocardiography (ECG) [25], electroencephalography (EEG) [26], photoplethysmography (PPG) [27], galvanic skin response (GSR) [28], skin temperature (ST) [29], heart rate (HR) [30], heart rate variability (HRV) [31], blood volume pressure (BVP) [32], electromyogrpahy [33] and by fusion of these bio-signals [34]. In literature, several acute human stress measurement studies using different stimuli have been proposed. The framework used for the assessment of human stress in the literature follows the following steps. Firstly, physiological signals are acquired by inducing stress either in laboratory settings [35-38] or in real life situations [39-43]. Followed by this, a discriminating set of features are extracted to classify human stress by either using some pattern recognition or machine learning techniques. Subsequently, a diverse range of machine learning techniques has been used to classify human stress into two and three levels. The public speaking activity has been used as a stress-inducing stimulus in various stress measurement studies [44,45]. An acute stress measurement scheme using unobtrusive physiological markers of skin temperature, heart rate, and pulse wave in response to the mental arithmetic task and public speaking stimuli was presented in Ref. [11]. The study concluded that the rest state (27.86 auxiliary units) can be discriminated from stress condition (47.55 auxiliary units) using these unobtrusive physiological bio-markers. Another human stress measurement study using HRV signals was presented in Ref. [46] and data was annotated using scores of the STAI questionnaire. A human stress assessment scheme using GSR and PPG signals in response to stimuli of cognitive task and public speaking was presented in Ref. [47]. Human stress assessment using physiological signals (GSR, PPG, and HRV) and sociometric sensors (microphone, accelerometer, proximity, and infrared sensors) in response to a cognitive task and public speaking stimuli were presented in Ref. [48]. Another study for analyzing the real-life public speaking anxiety using recurrence quantification analysis of the GSR signals was presented in Ref. [49]. Non-linear features of the GSR signals and neural network were used to identify the calm and the anxious state as well as low anxiety and high anxiety state. Temporal parametric models to quantify the bio-behavioral trajectories of public speaking anxiety using real-life public speaking stimuli and physiological signals of GSR, PPG, and HR was presented in Ref. [50]. The study concluded that the proposed time-based analysis of the bio-behavioral measures was more robust in estimating the characteristics of the participants as compared to the traditional model. The multimodal human stress assessment methods have remarkable results in terms of obtained accuracy as compared to single modality based methods [51,52]. Multimodal stress assessment schemes in response to different stimuli have been proposed in a wide range of studies available in the literature. Human stress measurement using a fusion of near-infrared spectroscopy (NIRS) and physiological signals of PPG, ECG, and seismocardiogram in response to mental arithmetic and memory tasks was presented in Ref. [53]. An accuracy of 85% was achieved for three states using a random forest (RF) classifier. Assessment of mental stress using a fusion of EEG and functional-NIRS signals in response to mental arithmetic task with an accuracy of 95.1% using SVM classifier was proposed in Ref. [54]. The fusion of EEG signals with other psychological modalities like ECG [55] and galvanic skin response [56] has been used for boosting the performance of stress classification. The fusion of GSR and ST has been used to classify human stress with an accuracy of 84% [57]. Heart rate, skin conductance, and accelerometer sensors have been adopted for the measurement of real-time stress recognition using daily-life activities [58]. An accuracy of 92.15% was achieved for three-level stress classification using the fusion of features from these three modalities. A novel multimodal stress recognition scheme using a fusion of features from BVP, HR, ST, GSR, and respiratory rate (RR) in response to the mental arithmetic task has been proposed in Ref. [59]. An accuracy of 83% and 72% is achieved for two and three class stress classification, respectively. Some of the other stress measurement studies available in the literature include fusion of ECG and contextual data obtained from smartphones [60], a fusion of accelerometer and electrodermal activity (EDA) data [61], a fusion of respiratory, ECG and accelerometer sensors data [62] and fusion of accelerometer, EDA and Bluetooth sensors data [63].🏁