Diseño de un sistema basado en tecnologías móviles, wearables y análisis de datos para promover el envejecimiento activo en mayores

Abstract

Traditional assessments of the elderly’s health status, such as frailty or dependence, rely on subjective responses to questionnaires. These questionaries have associated resource costs for healthcare systems and entail a lack of realistic prevention management. Hence, it is necessary to develop new technological systems capable of evaluating these health conditions, in an objective, ubiquitous and transparent way for the users, contributing to reducing the derived costs and promoting prevention. This work presents a mobile-Health system that collects and analyzes physiological data from the elders using various mobile devices. The system architecture follows the Internet of Things (IoT) paradigm, m-IoTHealth, interconnecting different mobile devices. Specifically, the system adopts the microservices architectural style, in which each component specializes in a single, fine-grained, task. These tasks consist of collecting, federating, and analyzing the heterogeneous data sources. To assess our system, we generated machine-learning models to evaluate the fragility and dependency status. We also proposed a motor imagery detection model as the first approach to study cognitive data. The results of this thesis provide a comprehensive technological solution to assess frailty and dependency of elderly people. This solution includes the high-level design of a microservices-based architecture with highly specialized components for the different tasks expected from an m-Health. To the best of our knowledge, this is the first time a system collects and processes physiological data generating machine-learning models during the performance of a complex activity (multidimensional: physical, social, and cognitive) such as shopping. The results are objective and accurate in the assessment of the frailty and dependency status of older adults. Thanks to the proposed design based on specialized microservices, the solution is flexible, and allows the reuse and integration of its different components to generate new assessment models, as in the case of our motor imagery detection model The generation of these models involved the use of different sensors embedded in the Samsung Gear S3, Empatica E4, and Muse 2 wearables, such as the accelerometer, gyroscope, heart rate, temperature, electrodermal skin activity (EDA), and electroencephalogram (EEG) signals. We use different preprocessing techniques, such as window segmentation and signal alignment, and explore several machine learning algorithms such as k-Nearest Neighbors, Random Forest, Naïve Bayes, and neural networks for time series (LSTM). To optimize the models, we search for hyperparameters that reduce dimensionality and identify relevant features. This proposal could promote the prevention of the deterioration of the health status by the early detection of decay. In particular, it could prevent frailty and dependence, reducing, at the same time, social and healthcare costs.