Prediction of Stress Levels using Low-Cost IoT-Based Health Parameters Measuring System

Human stress is a prime concern of health nowadays. Sometimes it affects the body state severely. When it is necessary to find or identify the stress level of anybody, patient or elderly person, continuous monitoring is required. So a low-cost system is required which can be used to measure different health parameters. For this, an IoT-based system has been developed to measure a few physical parameters such as temperature, pulse, and oxygen saturation (SpO2) using MAX30102 Pulse Oximeter and Heart Rate sensor which is subsequently interfaced with a node microcontroller unit for further processing. Whenever data is measured and collected using the implemented device, it has been observed that there are four types of people with stress levels: healthy, moderately stressed, highly stressed, and overstressed. Based on the received unknown data, several groups or clusters with similar data points and other groups or clusters with other dissimilar data points are formed using clustering-based approaches such as hierarchical-based agglomerative clustering, fuzzy-based fuzzy C-Means clustering, and centroid-based K-Means clustering. The selection of the number clusters is done by calculating the scores using the elbow method and the silhouette method. The obtained number of clusters is five using both methods, and this value has been kept the same for all clustering techniques. After this, the silhouette score is calculated for each clustering methods, and obtained the best score (0.693) from K-Means. Finally, the plot of means of each attribute has been analyzed for different cluster levels in each clustering algorithm.