앙상블 머신러닝 모델을 활용한 치매 예측

Abstract

As we become an aging society, there is a growing concern about geriatric diseases. Among them, senile dementia, which disrupts daily life, is best detected at an early stage because there is no cure, even with technological advances. However, many preliminary studies to predict dementia require continuous imaging data, such as MRI and retinal imaging data, and are expensive. These methods have the disadvantage that they cannot be applied to all people, so this study aims to develop low-cost predictive dementia detection. Therefore, the purpose of this study is to identify the main variables that most influenced the presence of dementia using deep learning prediction models such as XGBsoot and lightGBM by utilizing various activity data such as exercise duration, high-intensity activity duration, and metabolic rate, and sleep data such as deep sleep duration and breathing rate per minute to determine sleep quality. The analysis showed that the best dementia prediction performance was achieved when both activity and sleep data were considered simultaneously, and the most important variable in the XGBoost model was the “activity_score_move_every_hour” and the “sleep_hr_everage” in lightGBM.