Modelos de Previsão de Qualidade de Vida Através de Sensores Biométricos

• Autor(a) principal: Anaís Silva Dias
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/106900
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

With the improvement of health care and living conditions, there has been a decrease in mortality, accompanied by an increase in life expectancy of the overall population. This increase is also associated with a rise in chronic illnesses, which can induce a great deal of pain and discomfort. Therefore, it is important to not only focus on the length of a patient's life, but also on its quality. Even though Quality of Life (QoL) is now a crucial measurement in the treatment of patients, the methods to assess it are not as evolved as necessary. There is a lack of continuous evaluation, which is the most effective way of assessing QoL. Stemming from the need to predict and monitor QoL with greater precision, we aim to develop predictive models based on physiological data obtained through biometric sensors, by simulating a wide variety of this data and attributing QoL scores to it, depending on their features, and then training models with the resulting data. This will be done within the scope of the QVida+ research project, which aims to create an information system that will use physiological data from patients to evaluate QoL, and that will adapt continuously to the patient and serve as a decision support system for health care professionals. In order to achieve the best results, several biometric measurements will be analyzed. We will have in account both the precision of their output, but also their practicality for the patient, due to the need for constant wear. A large set of physiological data will be generated, with a great variation, in order to allow for extensive testing and tuning of the predictive models. This data will be labeled in terms of QoL, according to its features. Then, we will compare the possible methods of classification of the data, in order to achieve predictive models that will be precise and generalizable to a larger population, and that will allow a continuous measurement without significant effort required from the patient. Some of the techniques that will be tested in this step are random forests and deep learning. The possibility of assessing QoL continuously can have a significant impact in health care, serving as a support to clinical decision, as it provides a highly important and reliable measurement, benefiting patients and professionals alike.