Fraility estimation with machine learning and time scaleeffect

With the aging world population, early detection of frailty in elderly individuals plays a crucial role in enhancing the effectiveness of healthcare services and preserving individuals' quality of life. Frailty is characterized by declines in physical, cognitive, and psychological functions, which limit an individual's capacity for independent living. This study focuses on using machine learning algorithms to determine and predict frailty levels in elderly individuals based on biophysical data obtained from wearable devices. While most studies in the literature rely on clinical observation and tests, the continuous data flow from wearable devices enables more comprehensive and real-time analyses. In this thesis, accelerometer and ECG sensor data will be analyzed using algorithms such as XGBoost, KNN, RF, 1D-CNN, MLP, and Naive Bayes to identify the most suitable model and time window for frailty prediction. The research questions of this study address whether frailty can be accurately predicted using biophysical data, what the most effective time window is, and which algorithm performs best. Structured in five main chapters, this thesis covers a review of the relevant literature, data collection and analysis methods, algorithm applications, and findings of the research.