Percorrer por autor "Pina, Eduardo Manuel Abreu"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Machine Learning Models for Detecting Chronic Disease Progression Based on VO2max: A Study Applied to Cardiovascular Diseases and Type 2 DiabetesPublication . Pina, Eduardo Manuel Abreu; Duarte, Rui Pedro Monteiro AmaroChronic diseases, namely cardiovascular diseases and type 2 diabetes, remain the leading causes of mortality and morbidity around the world. The early detection and prevention of these conditions are crucial to improve health outcomes and reduce the burden on healthcare systems. Among existing health biomarkers, VO2max is a key feature traditionally used to evaluate individual performance during high in tensity activities. Moreover, there have been associations between VO2max levels and the presence of CVD and T2D. The appearance of wearable devices, such as smartwatches, enables the measurement of VO2max and other biomarkers, in a non invasive way, providing real time data for a reliable and validated set of information. The data obtained allows the implementation of machine learning models capable of predicting patterns and chronic disease progression. To this end, the project acqui sition of data using a Garmin Venu 2 Plus smartwatch, led to the creation of four datasets, with the help of OpenAI ChatGPT, due to the limited real world data collected for CVD and T2D assessment. These datasets comprised of data derived from the smartwatch and fully synthetic data to be benchmarked under different risk distributions. The feature selection technique using Recursive Feature Elimination identified VO2max, heart rate variability, sleep time, body mass index, respiration metrics, oxygen saturation and activity features as the most relevant features. The application of two supervised machine learning classification algorithms, SVM and LR, are evaluated, with SVM outperforming LR across both diseases. Notably, al though models trained on fully synthetic data showed a strong training performance, datasets derived from smartwatch data demonstrated a better generalisation dur ing testing. Overall, the results confirm the validity and effectiveness of combining non-invasive features with the application of machine learning models for chronic disease risk assessment, highlighting the importance of the application of realistic physiological data in predictive health monitoring systems.