Health monitoring via wearable sensors

With the development of technology, the collection, storage and processing of data has become widespread, and therefore wearable sensor technology has become increasingly popular in our daily life. As wearable sensors and smart textiles become smaller, cheaper and easily accessible by the user, they are widely used in daily activity tracking and healthcare. Wearable sensor technologies are used today as an effective tool for disease prevention, early disease detection and management of chronic conditions. The main purpose of using wearable sensors in the field of healthcare is to monitor the health parameters of patients remotely, and therefore to enable patients to be treated at home and prognose the disease status without having to go to a healthcare center. To achieve a baseline in this manner, experiments were carried out in this thesis using the signals obtained with wearable sensors to prognose diseases. For this purpose, a fatal infectious disease called Sepsis and a neurological disease without treatment called Parkinson's disease, were selected for thesis study. Vital signs that can be easily obtained with wearable sensors are used in Sepsis prognose experiments. These vital signs have been used to predict the organ failure score that allows monitoring the disease status. Parkinson's disease symptom level value estimation was performed by walking analysis with wearable shoe system that measures the force applied to the foot. In order to perform these predictions, CNN and Random Forest based deep learning hybrid architectures are developed. Also, experiments are conducted with traditional machine learning and deep learning architectures to validate the performance of this approach. In experimental results, it was observed that the developed hybrid architectures provide performance increase compared to traditional methods.