應用最鄰近區域分類法於慢性病分類預警準則之研究

Using K-Nearest Neighbor Classification to Establish Chronic Illness Early Warning Criterion

朱晏嬅(Yan-Hua Chu)

指導教授 ： 江行全

元智大學/工程學院/工業工程與管理學系/碩士(2008年)

https://doi.org/10.6838/YZU.2008.00169

摘要

近年來慢性病逐漸取代急性傳染病，成為國人主要死因之一，且罹患慢性病容易併發其他相關病症，如高血壓是臨床上最常見的一種指標症狀。故本研究著重於預防的角度，期望找出影響多種慢性疾病的重要生理指標與臨床檢驗值，且利用這些危險因子架構多種慢性病的分類預警準則，提供健康民眾或是慢性病患在病症發作前察覺其預兆，降低多種慢性病的併發率。

本研究是以北部某醫學中心的健檢資料作為研究資料來源，主要探討高血壓、糖尿病、心血管疾病、肝病、腎臟病以及無罹患上述五種慢性病的正常人。利用最鄰近區域分類法、線性鑑別式分析與逐次前饋式搜尋法將本研究分為二大部分進行探討：第一部份針對正常人與各種慢性疾病進行分類與特徵值篩選；第二部份則是找出各個慢性病的危險因子之臨界值並建構多種慢性病的分類預警準則。

實驗結果顯示針對健檢資料的分類與危險因子篩選具有良好效果，分類正確率最高可達98%，且透過這些重要因子所架構的分類預警準則可協助病患了解自身罹病的風險，亦可有效地提供醫療人員輔助診斷之參考。

關鍵字

慢性病 ； 最鄰近區域分類法 ； 線性鑑別式分析 ； 逐次前饋式搜尋法 ； 分類預警準則

並列摘要

Chronic disease has gradually displaced persistent infectious disease as one of the major causes of death in Taiwan. Being afflicted with such illnesses elevates vulnerability to other syndromes as well. For instance, high blood pressure is a common indicator symptom cited in clinical records. Thus, this research adopts a preventative perspective and attempts to ascertain the impacts of important physiological indicators and clinical examination values for various chronic illnesses. Utilizing dangerous factors from a variety of chronic diseases to establish an early warning criterion may reduce the syndrome occurrence rate. Criteria offer an early warning to healthy people or chronic patients to aid them in perceiving the signs of illness before symptoms emerge.

The research samples primarily represent five chronic diseases: high blood pressure, diabetes, cardiovascular disease, disease of the liver, diseases of the kidneys. They were collected from a medical center and compared with the data of healthy people. K-Nearest Neighbor (K-NN) classification, Linear Discriminate Analysis (LDA), and Sequential Forward Selection (SFS) are utilized in the research, which is divided into two parts. The first part classifies and screens both healthy persons and those afflicted with the abovementioned chronic illnesses for characteristic value determination, and the second part determines the critical value of the dangerous factors of each chronic illness and builds early warning criteria to categorize the chronic illnesses.

The results reveal that classifying materials and screening dangerous factors are both positively efficient with a corrected rate of 98%. Additionally, through the important factors of early warning criteria, not only can we help patients understand the risks of suffering from such diseases, but also effectively offer superior diagnosis reference criteria for medical personnel.

並列關鍵字

Chronic Illness ； K-Nearest Neighbor(K-NN) ； Linear Discriminant Analysis(LDA) ； Sequential Forward Selection(SFS) ； Early Warning Criterion

參考文獻

姚志成 (2005)，「運用資料探勘技術建構脂肪肝預測模式」，碩士論文，中原大學資訊管理研究所。

陳筱筠、黃麗卿 (2005)，「某職場空腹血糖異常與糖尿病發生率及其相關因素之研究」，台灣家醫誌，15卷4期，pp. 191-201。

李博智 (2002)，「資料探勘在慢性病預測模式之建構」，碩士論文，元智大學資訊管理研究所。

張志豪 (2005)，「強健性和鑑別力語音特徵擷取技術於大詞彙連續語音辨識之研究」，碩士論文，臺灣師範大學資訊工程研究所。

黃盈禎 (2006)，「探討早期慢性腎疾病患者身體活動量、體能與腎功能之關係」，碩士論文，成功大學護理學系成人護理組。