웨어러블 기기 라이프로그를 활용한 치매 진단 예측 머신러닝 모델 비교
Comparison of Machine Learning Models for Predicting Dementia Diagnosis Using Wearable Device Lifelog Data
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
In response to rapid population aging in South Korea, this study aims to develop and validate an accurate, non-invasive dementia screening tool by comparing and evaluating five machine learning models. We collected wearable device lifelog time-series data—including sleep and gait patterns—and cognitive test results from target patients, then preprocessed these inputs for LSTM, bi-LSTM, Random Forest, XGBoost, and 1D-CNN classifiers. Performance was assessed by accuracy, ROC-AUC, and F1-score; the 1D-CNN outperformed the others with 78.8% accuracy, 81.0% AUC, and a 0.63 F1-score. Logistic regression applied to combined lifelog and cognitive variables identified sleep breathing rate, sleep movement rate, and immediate recall ability as significant predictors. These findings suggest that machine learning models can support early and continuous dementia risk monitoring and inform tailored intervention strategies.
Abstract
In response to rapid population aging in South Korea, this study aims to develop and validate an accurate, non-invasive dementia screening tool by comparing and evaluating five machine learning models. We collected wearable device lifelog time-series data—including sleep and gait patterns—and cognitive test results from target patients, then preprocessed these inputs for LSTM, bi-LSTM, Random Forest, XGBoost, and 1D-CNN classifiers. Performance was assessed by accuracy, ROC-AUC, and F1-score; the 1D-CNN outperformed the others with 78.8% accuracy, 81.0% AUC, and a 0.63 F1-score. Logistic regression applied to combined lifelog and cognitive variables identified sleep breathing rate, sleep movement rate, and immediate recall ability as significant predictors. These findings suggest that machine learning models can support early and continuous dementia risk monitoring and inform tailored intervention strategies.