Long-Term Prediction of Continuous Health Measurements
Name
Belinda Lepmets
Abstract
This bachelor’s thesis explores the possibility and methodology of creating a model for predicting continuous health measurements over a long-term period using short-term health data. To achieve this, a similarity-based prediction model was developed, relying on a short-term observation window dataset containing measurements from numerous patients. The most similar patient is selected based on initial parameters, and step-by-step progression through their and other patients’ data forms the basis for long-term prognosis. This model was demonstrated using four test cases: height, weight, pulse, and systolic blood pressure, utilizing data from the RITA-MAITT and ELIKTU studies. The height prediction model showed the highest predictive accuracy among these test cases, with an average error of 5.5 cm for model testing and 5.1 cm for the “average of three predictions” model. While this approach may not be suitable for predicting all continuous health measurements, it yields sufficiently accurate results for certain indicators, thus warranting further refinement and investigation of the method.
Graduation Thesis language
Estonian
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Sulev Reisberg
Defence year
2024