Conceptual design of a machine learning-based wearable soft sensor for non-invasive cardiovascular risk assessment

The number of elderly people is increasing, and heart diseases are a major issue in a healthy aging of population. Indeed, the possibility of hospital care is limited and the avoidance of crowded hospitals recently became even more essential. Meanwhile, the possibility to exploit e-health technology for home care would be desirable. In this framework, the concept design of a soft sensor for measuring cardiovascular risk of a patient in real time is here reported. ECG, blood oxygenation, body temperature, and data acquired from patients' interviews are processed to extract characterizing features. These are then classified to assess the cardiovascular risk. Experimental results show that patients' classification accuracy can be as high as 80% when employing a random forest classifier, even with few data employed for training. Finally, method evaluation was extended by exploiting further data and by means of a noise robustness test.