- 學位論文
智慧型基因演算法於疾病分類規則之研究
A Study of Rules Discovery For Disease Classification Using Intelligent Genetic Algorithms
摘要
病歷電子化作業的推廣後，醫學診斷資料也正式的進入了全面電子化的時代，不但可節省廣大的病歷儲存空間，也促進了醫療機構之間彼此的醫療資源整合。而在廣大的醫療診斷資料中，其實隱含了許多規則和知識，且這些規則和知識可能是之前所未發現的；這將可對預防醫學上提供一個潛在的發展方向。 資料探勘技術可在廣大的資料中，發現隱藏的知識。但諸多的資料探勘技術，大都屬於黑盒技術，無法直接成為外顯的知識規則。而由資料探勘進而提升至知識發現，在電子病歷中萃取出有用的知識及規則；透過這種機制，可提供相關的醫療人員進行知識的驗證與後續知識分享的應用 本研究提出智慧型基因演算法（Intelligent Genetic Algorithms, IGAs），結合了實驗設計與適應性突變機制，改善了傳統實數型基因演算法容易陷入局部極值、及演化時間耗時的問題，經4組驗證實驗共56個測試函數評估後，證實其求解精確度確優於前人提出之諸多演算法。 最後將本研究研發的IGAs再延伸開發成智慧知識規則萃取系統(Intelligent Knowledge Rules Extraction System, IKRES)，採用UCI(University of California at Irvine)疾病案例資料集，進行疾病分類規則萃取實驗；在四個二元分類及二個多元分類的實際病例分類結果顯示，本研究研發的知識規則萃取系統比前人的方法，具有更高的準確率與更低的偽陰性誤診率。
並列摘要
Due to the promotion of electronic medical records, medical diagnosis data also entered the era of fully electronic. The electronic medical records not only save storage space, but also integrate resources between different medical institutions. There is a lot of rules and knowledge hidden in large amounts of medical diagnosis data, which may be previously undiscovered and can provide a potential development direction for preventive medicine. Data mining technology can discover hidden knowledge in the vast database. But most of data mining techniques are black-box approaches which can not directly be explicit rules of knowledge. Knowledge discovery in data mining can directly extract the useful knowledge rules from the electronic medical records. The extracted knowledge rules can be verified by the relevant medical staff and then can be applied through knowledge sharing. At first, this study proposed an intelligent real-coded genetic algorithm (IRGA) to improve the performance of traditional real-coded genetic algorithms by using experimental design based crossover and the adaptive mutation mechanism. Comparing with the previous proposed algorithms, the simulated results of 56 benchmark function optimization experiments clearly demonstrated the superiority of the proposed IRGA method. Finally, this study extended the IRGA to develop an intelligent knowledge rules extraction system (IKRES) for automatic disease classification rule extraction. By using the UC Irvine Machine Learning Repository, 6 real-life illness datasets including 4 binary classification cases and 2 multi-category classification cases had investigated in this study. Comparison with the previous rule extraction systems, the results evidently showed that the proposed IKRES method with higher accuracy and lower false negative rate of misdiagnosis.