Stress prediction using internet of things-based galvanic skin response measurement

Early detection of stress is important for the protection of mental and physical health. Physiological data collection and analysis methods with wearable devices are becoming widespread. Machine and deep learning techniques are used to estimate stress levels by evaluating biometric parameters and physiological signals. In this study, the success of different machine and deep learning algorithms in stress estimation was compared on an open dataset. The dataset containing physiological signals such as heart rate, galvanic skin response, respiratory rate and temperature was analysed with machine learning algorithms such as gradient descent decision tree, random forest and support vector machine. The results showed that the support vector machine with radial basis function kernel performed better than other algorithms. In addition, deep learning algorithms such as recurrent neural network, long short-term memory, gate recurrent cells and hybrid model with attention layer were tested. As a result of the performance evaluation, it was seen that the proposed hybrid model with attention layer was superior to other methods in terms of mean absolute error metric in stress estimation. The results of the study reveal that stress monitoring is possible with wearable devices and that this technology can be integrated into individual health management systems. In future studies, it is aimed to apply different deep learning techniques optimized using larger and more original data sets and to increase stress prediction performance by integrating multi-modal biophysiological data. In this way, it is anticipated that current studies can be carried further.