Resp-BoostNet: Mental Stress Detection From Biomarkers Measurable by Smartwatches Using Boosting Neural Network Technique

To maintain overall health and well-being, it is crucial to manage mental stress. This study focuses on developing a deep learning model for recognizing mental stress levels using the sensors of smartwatches. Most related research with notable performance has focused on mental stress detection using various physiological biomarkers obtained through sophisticated IoMT (Internet of Medical Things) devices. However, the ones using only the smartwatch’s measurable physiological biomarkers, which do not include respiration rate, have comparatively lower performance because of a limited number of physiological biomarkers. In this paper, we introduce an improved model for mental stress detection using boosting neural network that can be integrated into a smartwatch. The proposed model consists of two phases. In the first phase, we introduce a boosting neural network technique that predicts the respiration rate by utilizing the biomarkers measurable by a smartwatch. In the second phase, the modified set of biomarkers, which includes both the original biomarkers and the predicted respiration rate, is used for stress level classification via an artificial neural network. The necessary hyperparameter tuning is performed to obtain the optimal values of various model parameters. The training of the model is performed for fifteen different subjects of the publicly available multimodal WESAD (Wearable Stress and Affect Detection) dataset using various biomarkers measured by smartwatches. The proposed model predicts respiration rate with low error (0.035 MSE (Mean Squared Error)) and achieves high mental stress detection accuracy of 94% using smartwatch measurable biomarkers which is a ~2% improvement over the current contemporary technique.