ScholarWorks@성균관대학교

초록

Cerebrovascular disease (CeVD) is a major cause of mortality and disability, necessitating continuous monitoring for timely intervention. Home environments are effective for contactless health monitoring, as CeVD risk factors can be continuously tracked without obtrusion. This study leveraged smart home data and artificial intelligence (AI) to predict CeVD emergencies early. The dataset included individual health conditions and sequential life-logs, such as physical activity, sleep, and thermal environment, from 1130 CeVD patients, including 130 emergencies. We achieved an area under the precision-recall curve (AUPRC) of 0.94 with 28 days of data, corresponding to an accuracy of 98.2 %, precision of 97.1 %, recall of 87.2 %, and an F1-score of 91.9 %, using the CrossNet architecture, which integrates static and time-series data while handling missing values by cross-modal imputation. We further identified emergency prevention strategies, including: increasing both low and high active time by 1 and 1.5 h/day, while decreasing inactive time for patients aged 85+; limiting high active time while increasing low active time for patients aged under 85; maintaining 7 h of sleep for cardiovascular patients (8 h if cardiovascular); minimizing sleep fragmentation for patients aged 85+ and with diabetes; and in general, cold indoor temperatures increase the emergency risk, while hot indoor temperatures are risky in cold weather. These findings highlight the potential of smart home monitoring based on the artificial intelligence of things (AIoT) to predict emergencies and identify prevention strategies. This study provides insights into scalable, contactless, AI-enabled home healthcare solutions for continuous management of CeVD.

키워드