IoT and machine learning enabled estimation of health indicators from ambient data

Physiological health indicators can provide valuable insights into the general health and well-being of a person. However, acquiring these indicators implies being physically connected to a medical device or using wearable sensors. Moreover, the aforementioned devices only measure the indicators but provide no information on what influences them. This study proposes an approach for estimating such indicators from ambient data, enabling simultaneously non-invasive monitoring and providing details on how the environment affects one's health. A system based on Internet of Things (IoT) sensors is used for data collection and Machine Learning (ML) algorithms are employed for data analysis. The study focused on two health signals, Heart Rate (HR) and Skin Resistance (SR). Out of the three tested algorithms, Random Forest (RF) yielded the best results in terms of Mean Absolute Error (MAE) for both indicators. The results obtained proved that physiological signals estimation exclusively from ambient data is possible and identified which environmental factors are most important.