Using Accelerometer Data and Machine Learning as Support for Clinical Studies

Recent improvements in technologies used in the medical & health field have significantly supported specialists in analyzing and studying diseases. Detecting daily life activities with the help of wearable devices might become a precise and objective method for remote patient monitoring. The goal of this thesis is to investigate the use of data obtained from wearable devices and the application of Machine Learning methods as support for clinical studies. Our primary task is to classify activities of daily life focusing, in our case, on using accelerometer data. Thanks to the study of the state-of-the-art, we noted three main areas in which to contribute: (1) public datasets, (2) methodology, and (3) reproducibility. Regarding the first area, our contribution has been to record, prepare, and share two datasets of accelerometer data including daily life activities. The first dataset was recorded with healthy adults, the second involves children and adolescents suffering from chronic diseases, such as Juvenile Fibromyalgia, Juvenile Idiopathic Arthritis, and Juvenile Dermatomyositis. Concerning the second area, our main effort has been in proposing a method that aids the classification of non-interesting activities for specialists, reducing the impact on the classification of known activities, and testing different approaches to improve this outcome. With respect to the third area, we devised a review that analyzes the situation of related works, with a focus on sharing the used datasets and biometric information concerning the subjects included in the studies.