根據世界衛生組織指出，2016年全球的慢性阻塞性肺病（Chronic Obstructive Pulmonary Disease，COPD）病例達到2.51億例，為目前全球第三大死因。美國國家心臟、肺和血液研究所指出起初COPD可能不會有任何症狀，且與其他疾病容易混淆，使得患者常常未能在第一時間得到適切的診斷，延誤治療的時機。
現今COPD臨床診斷流程是依患者症狀、病因再加上肺功能檢查來綜合評估診斷，但因患者之間的個別健康與病因狀況差異很大，且目前對於COPD的病因仍未清楚了解，於是有了評估性問卷的方式來輔助診斷。目前有許多研究指出使用評估性問卷，更確實可以在一般情況下發現疑似COPD患者，提高當前診斷準確性以期能達精準醫療（precision medicine）之成效，可惜的是目前模型準確度與後續實際即時臨床診斷系統服務之應用似乎略顯不足。
因此，本研究發展出利用美國疾病管制與預防中心（Center of Disease Control and Prevention，CDC）危險因子監控系統（Behavioral Risk Factor Surveillance System，BRFSS）的資料建立一套機器學習預測模型，除了找出影響COPD預測的關鍵因素外，並應用SHAP（Shapley additive explanations）技術來分析解釋特徵是如何影響最終模型預測結果，最後以最佳模型為基礎，建置COPD輔助診斷系統平台。本系統提供視覺化輔助診斷結果報告，能直接查看每項特徵風險的權重大小，使用者能以視覺方式理解每項特徵因素對預測的影響，並藉以輔助醫師診斷或提供民眾能自行判斷是否需要就醫，因而能早日取得治療契機，提高國家醫療資源效能，減輕家庭與長期照護之負擔。

According to the World Health Organization (WHO), 251 million cases of Chronic Obstructive Pulmonary Disease (COPD) occurred worldwide in 2016, making it the world's third leading cause of death. According to the National Heart, Lung, and Blood Institute, the COPD may not present any significant symptoms, thus many patients may not be confirmed at the beginning lead to delaying treatment.
Nowadays the confirmation of COPD needed following clinical symptoms, etiology and lung function. However, due to the differences in individual health among patients, and the causes of COPD are still not clearly understood at present, the evaluative questionnaire is adopted to assist the diagnosis. At present, many studies have pointed out that the use of an evaluative questionnaire can find suspected COPD patients in general conditions effectively, it improves the current diagnostic accuracy and achieves the effect of precision medicine. Unfortunately, the application of a combination of the accuracy model and the clinical diagnosis system services seems insufficient.
Therefore, this study employed several Machine Learning predictions to find a best fit COPD prediction model based on the data of the Behavioral Risk Factor Surveillance System (BRFSS) of the Center for Disease Control and Prevention (CDC). In addition to identifying the key factors affecting the prediction of COPD, Application SHAP (Shapley additive explanations) technology is used to explain the impact of features in final model-predicted results. Finally, based on the best model, applied to build a COPD web-based diagnosis-aid system platform, the system provides visual supporting diagnosis message, it can measure the weight of each risk characteristics, the user can visually realize each characteristic factor's influence on the prediction, thus may assist in physician diagnosis and provide people self-evaluation the necessity to go to the doctor for early treatment. Therefore, promote the efficiency of national health resources and alleviate the family’s financial burden in long-term care.

誌 謝 I
摘 要 II
Abstract IV
目 錄 VI
圖目錄 VIII
表目錄 X
第壹章 緒論 1
第一節 研究背景 1
第二節 研究動機 3
第三節 研究目的 4
第四節 研究流程 4
第五節 重要名詞解釋 6
第貳章 文獻探討 8
第一節 COPD定義與介紹 8
第二節 行為危險因子監控系統 (BRFSS) 17
第三節 人工智慧與機器學習 17
第四節 模型評估與選擇 24
第參章 研究方法 30
第一節 研究工具 31
第二節 資料來源 31
第三節 變項挑選 32
第四節 資料預處理 33
第五節 機器學習模型訓練與評估 38
第六節 輔助診斷系統平台建置 43
第肆章 研究結果 45
第一節 統計分析 45
第二節 特徵選擇 48
第三節 模型結果 49
第四節 輔助診斷系統平台 59
第伍章 討論 63
第一節 模型比較與特徵探討 63
第二節 本研究之應用性 69
第三節 研究限制 69
第陸章 結論 70
第一節 研究結論 70
第二節 研究貢獻 71
第三節 未來研究方向與研究建議 71
參考文獻 72
附錄 76
附錄一 2018 BRFSS變項問題 76
附錄二 COPD輔助診斷系統平台 80

中文部分
[3] 簡郁芬、葉昱麟、張富勝，慢性阻塞性肺病的疾病分期與治療新觀念，家庭醫學與基層醫療，第28卷，第5期，頁121-127，2013。
[5] 衛生福利部國民健康署，108年度死因統計- 統計處，網址： https://dep.mohw.gov.tw/DOS/lp-4927-113.html，上網日期：2020年7月1日。
[6] 余忠仁、黃忠智、林慶雄 等，台灣肺阻塞照護精要「全民健康保險肺阻塞醫療給付改善方案」教育訓練核心教材，台灣胸腔暨重症加護醫學會，ISBN：978-986-92595-8-3，106。
[7] 蘇湘雲，肺阻塞 7成沒聽過，網址：http://asthma-copd.tw/lectures1_detail.php?id=18，上網日期：2020年5月10日。
[8] 杜杏慧，胸部電腦斷層篩檢慢性阻塞性肺病之正面效益，防癆雜誌，頁16-17，2013。
[11] 慢性病防治委員會，慢性阻塞性肺病自我管理手冊，慢性病防治委員會出版，2014。
[12] 張芬菱，慢性阻塞性肺病COPD，防癆雜誌，頁18-23，2014。
[13] 台灣肺阻塞臨床照護指引編輯小組，台灣肺阻塞臨床照護指引，台灣胸腔暨重症加護醫學會出版，2019。
[14] 蘇一峰，[2020 GOLD Reports] 2020 COPD的黃金治療指引：重點整理-Part II，網址：http://tvgh-suvy.blogspot.com/2019/11/2020-gold-reports-2020-copd-part-ii.html，上網日期：2020年4月1日。
[15] 鐘威昇，慢性阻塞性肺疾病 2017 GOLD 的運用及省思，防癆雜誌，頁19-21，2017。
[46] 張紹勳，生物醫學統計：使用Stata分析，五南出版，2017。

英文部分
[1] World Health Organization. (2017). Chronic obstructive pulmonary disease (COPD). Accessed: 2020/3/10. Available: https://www.who.int/news-room/fact-sheets/detail/chronic-obstructive-pulmonary-disease-(copd)
[2] L. National Heart, and Blood Institute (NHLBI),. COPD (Also known as Chronic Obstructive Pulmonary Disease, Emphysema). Accessed: 2020/3/10. Available: https://www.nhlbi.nih.gov/health-topics/copd
[4] S. C. Lareau, B. Fahy, P. Meek, and A. Wang, "Chronic Obstructive Pulmonary Disease (COPD)," American journal of respiratory critical care medicine, vol. 199, no. 1, pp. P1-P2, 2019.
[9] B. Bratić, V. Kurbalija, M. Ivanović, I. Oder, and Z. Bosnić, "Machine learning for predicting cognitive diseases: methods, data sources and risk factors," Journal of medical systems, vol. 42, no. 12, p. 243, 2018.
[10] Global Initiative for Chronic Obstructive Lung Disease (GOLD), Global Strategy for the Diagnosis, Management, and Prevention for Chronic Obstructive Pulmonary Disease. 2020.
[16] Global Initiative for Asthma (GINA), Diagnosis of Diseases of Chronic Airflow Limitation: Asthma COPD and Asthma - COPD Overlap Syndrome (ACOS). 2014.
[17] E. Ray and C. Kelly, "The overlap between asthma and COPD: a case study," Independent Nurse, vol. 2019, no. 10, pp. 24-30, 2019.
[18] M. D. Eisner et al., "Socioeconomic status, race and COPD health outcomes," Journal of Epidemiology Community Health, vol. 65, no. 1, pp. 26-34, 2011.
[19] P. Lou et al., "Prevalence and correlations with depression, anxiety, and other features in outpatients with chronic obstructive pulmonary disease in China: a cross-sectional case control study," BMC pulmonary medicine, vol. 12, no. 1, p. 53, 2012.
[20] C. F. Vogelmeier et al., "Global strategy for the diagnosis, management, and prevention of chronic obstructive lung disease 2017 report. GOLD executive summary," American journal of respiratory critical care medicine, vol. 195, no. 5, pp. 557-582, 2017.
[21] V. Moore, "Spirometry: step by step," Breathe, vol. 8, no. 3, pp. 232-240, 2012.
[22] O. M. Mets et al., "Identification of chronic obstructive pulmonary disease in lung cancer screening computed tomographic scans," Jama, vol. 306, no. 16, pp. 1775-1781, 2011.
[23] D. Spathis and P. Vlamos, "Diagnosing asthma and chronic obstructive pulmonary disease with machine learning," Health informatics journal, vol. 25, no. 3, pp. 811-827, 2019.
[24] D. B. Price et al., "Symptom-based questionnaire for identifying COPD in smokers," Respiration, vol. 73, no. 3, pp. 285-295, 2006.
[25] N. Raghavan et al., "Components of the COPD Assessment Test (CAT) associated with a diagnosis of COPD in a random population sample," COPD: Journal of Chronic Obstructive Pulmonary Disease, vol. 9, no. 2, pp. 175-183, 2012.
[26] J. M. Guirguis-Blake, C. A. Senger, E. M. Webber, R. A. Mularski, and E. P. Whitlock, "Screening for chronic obstructive pulmonary disease: evidence report and systematic review for the US preventive services task force," Jama, vol. 315, no. 13, pp. 1378-1393, 2016.
[27] P. A. Frith, "Detecting COPD using micro‐spirometry and/or questionnaire," Respirology, 2019.
[28] D. Price, D. Freeman, J. Cleland, A. Kaplan, and F. Cerasoli, "Earlier diagnosis and earlier treatment of COPD in primary care," Primary Care Respiratory Journal, vol. 20, no. 1, pp. 15-22, 2011.
[29] National Center for Chronic Disease Prevention and Health Promotion Division of Population Health. CDC - About BRFSS. Accessed: 2020/3/5. Available: https://www.cdc.gov/brfss/about/index.htm
[30] D. E. Nelson, D. Holtzman, J. Bolen, C. A. Stanwyck, and K. A. Mack, "Reliability and validity of measures from the Behavioral Risk Factor Surveillance System (BRFSS)," Sozial-und Praventivmedizin, vol. 46, pp. S3-42, 2001.
[31] B. J. Copeland. Artificial intelligence. Accessed: 2020/8/5. Available: https://www.britannica.com/technology/artificial-intelligence
[32] S. A. Bini, "Artificial intelligence, machine learning, deep learning, and cognitive computing: what do these terms mean and how will they impact health care?," The Journal of arthroplasty, vol. 33, no. 8, pp. 2358-2361, 2018.
[33] Wikipedia. Supervised learning. Accessed: 2020/3/15. Available: https://en.wikipedia.org/wiki/Supervised_learning
[34] D. Lavanya and K. U. Rani, "Performance evaluation of decision tree classifiers on medical datasets," International Journal of Computer Applications, vol. 26, no. 4, pp. 1-4, 2011.
[35] Wikipedia. Decision tree. Accessed: 2020/3/1. Available: https://en.wikipedia.org/wiki/Decision_tree
[36] L. Breiman, "Random forests," Machine learning, vol. 45, no. 1, pp. 5-32, 2001.
[37] T. Chen and C. Guestrin, "Xgboost: A scalable tree boosting system," in Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, 2016, pp. 785-794.
[38] S. Ermon, C. Gomes, A. Sabharwal, and B. Selman, "Taming the curse of dimensionality: Discrete integration by hashing and optimization," in International Conference on Machine Learning, 2013, pp. 334-342.
[39] B. Sahu, S. Dehuri, and A. Jagadev, "A Study on the Relevance of Feature Selection Methods in Microarray Data," The Open Bioinformatics Journal, vol. 11, no. 1, 2018.
[40] S. Arlot and A. Celisse, "A survey of cross-validation procedures for model selection," Statistics surveys, vol. 4, pp. 40-79, 2010.
[41] Q. Ren, M. Li, and S. Han, "Tectonic discrimination of olivine in basalt using data mining techniques based on major elements: a comparative study from multiple perspectives," Big Earth Data, vol. 3, no. 1, pp. 8-25, 2019.
[42] A. C. Müller and S. Guido, Introduction to machine learning with Python: a guide for data scientists. " O'Reilly Media, Inc.", 2016.
[43] Wikipedia. Receiver operating characteristic. Accessed: 2020/4/2. Available: https://en.wikipedia.org/wiki/Receiver_operating_characteristic
[44] K. A. Spackman, "Signal detection theory: Valuable tools for evaluating inductive learning," in Proceedings of the sixth international workshop on Machine learning, 1989, pp. 160-163: Elsevier.
[45] M. H. Zweig and G. Campbell, "Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine," Clinical chemistry, vol. 39, no. 4, pp. 561-577, 1993.
[47] TIOBE. TIOBE Programming Community Index. Accessed: 2020/4/3. Available: https://www.tiobe.com/tiobe-index/
[48] A. Zheng and A. Casari, Feature engineering for machine learning: principles and techniques for data scientists. " O'Reilly Media, Inc.", 2018.
[49] W. McGinnis. Target Encoder — Category Encoders latest documentation. Accessed: 2020/5/8. Available: https://contrib.scikit-learn.org/categorical-encoding/targetencoder.html
[50] D. A. Cieslak and N. V. Chawla, "Learning decision trees for unbalanced data," in Joint European Conference on Machine Learning and Knowledge Discovery in Databases, 2008, pp. 241-256: Springer.
[51] A. More, "Survey of resampling techniques for improving classification performance in unbalanced datasets," arXiv preprint arXiv:.06048, 2016.
[52] S. M. Lundberg and S.-I. Lee, "A unified approach to interpreting model predictions," in Advances in neural information processing systems, 2017, pp. 4765-4774.
[53] S. M. Lundberg, G. G. Erion, and S.-I. Lee, "Consistent individualized feature attribution for tree ensembles," arXiv, 2018.
[54] S. Lundberg. slundberg/shap: A game theoretic approach to explain the output of any machine learning model. Accessed: 2020/6/20. Available: https://github.com/slundberg/shap
[55] Y. Ma et al., "Chronic obstructive pulmonary disease in rheumatoid arthritis: a systematic review and meta-analysis," Respiratory research, vol. 20, no. 1, p. 144, 2019.