| 摘要: | 隨著臺灣即將邁入超高齡社會，65歲以上老年人口持續攀升，其中獨居人口占比近四分之一，面臨居家安全、身心健康及社會支持等多重挑戰。跌倒為高齡者失能與住院的主要原因之一，若未及時獲得援助，可能造成臥床、併發症甚至死亡。為因應此風險，各地方政府推動緊急救援系統建置與在地老化照護服務，以強化獨居老人之安全保障。
本研究以新北市65歲以上獨居失能長者為研究對象，運用緊急救援系統偵測所得之活動數據與生理資訊，探索醫警事件（如跌倒）之發生機率與影響因子，並導入機器學習技術進行風險預測。研究採用監督式學習方法，包括Logistic Regression、Decision Tree、Random Forest、XGBoost與SVM等演算法，透過scikit-learn進行模型訓練與驗證。分析結果顯示Decision Tree模型在AUC、F1-score及MAE等指標上具最佳表現，具備高度解釋性與預測準確性，適合實務佈局。
本研究亦發現，個案之ADL/IADL、血壓值、疾病史與年齡等因子與醫警事件具有顯著關聯。未來可進一步結合時序資料與個人化特徵模型，提升預測精度與系統反應能力，提供智慧照護技術於社區長照服務應用之參考依據，以降低獨居長者居家事故風險，進而提升健康照護成效與生活品質。
;As Taiwan approaches a super-aged society, the proportion of individuals aged 65 and above continues to rise, with nearly one-fourth living alone. These elderly individuals face multiple challenges including home safety, declining health, and lack of social support. Falls are a leading cause of disability and hospitalization among seniors, and delayed assistance can result in severe consequences such as immobility, complications, or even death. To address these risks, local governments have introduced emergency assistance systems and developed community-based long-term care services aimed at enhancing safety and quality of life for older adults living alone.
This study focuses on elderly individuals aged 65 and over in New Taipei City who live alone and have installed subsidized emergency rescue systems. By analyzing activity data collected via the system, along with physiological indicators and functional assessment scales (ADL/IADL), the study identifies factors associated with medical alert incidents such as falls, and applies machine learning techniques to forecast risk. Supervised learning models including Logistic Regression, Decision Tree, Random Forest, XGBoost, and SVM were implemented and validated using scikit-learn. Among these, the Decision Tree model demonstrated superior performance in terms of AUC, F1-score, and MAE, making it a practical option for deployment in real-world settings.
The findings highlight significant correlations between variables such as daily functional capacity, blood pressure, chronic conditions, and the occurrence of emergency incidents. This research confirms the potential of artificial intelligence applications in predicting fall risks and provides actionable insights for optimizing elderly care systems. Future studies may integrate time-series data and personalized modeling to enhance precision and applicability across diverse living environments. |
