Mental Stress Prediction Using Wrist Wearable Through Machine Learning Approaches

Stress is one of the mental health issues that people face today. In reality stress is the primary cause of many conditions, such as heart disease, migraines, infertility, obesity, insomnia, and a host of others. In this scenario, stress prediction is crucial at an early stage so that we can protect ourselves against future issues. For identifying the stress, method of connecting sensors to multiple data gathering units using traditional technologies typically involved electrodes, wires, and sensors, was uncomfortable. Smart watches, on the other hand, are designed to be worn on the wrist and have sensors integrated into them that can measure heart rate and activity levels. Here, we report on our smart watch's highest level of stress prediction accuracy. We have used random forests, decision trees, support vector machines, naive bayes, logistic regression, and k-nearest neighbor models with k-fold cross-validation and voting ensemble learning to predict stress using data from smart watches (Fit bit). The support vector machine model had a binary class prediction of stress accuracy of 94°, which was the highest. Our findings suggest that the support vector machine and voting ensemble combination may be more effective at identifying stress using smart watches. Our study closes a variety of methodological shortcomings in previous studies on the automatic stress prediction using smart watches. It has also been noted that the model performs better when ensemble learning techniques are used.