台灣COVID-19本土日確診數之預測模式

Forecast models of the number of daily confirmed COVID-19 cases in Taiwan

巫岳霖(YUE-LIN WU)

指導教授 ： 陳慧芬

中原大學/電機資訊學院/工業與系統工程學系/碩士(2024年)

https://doi.org/10.6840/cycu202400167

摘要

由於COVD-19從2020年初開始在全球造成了經濟、健康、心理等諸多負面影響，而對於確診數、死亡數之預測模式的建立是一項重要工作，利用預測區間能夠及早的偵測爆發，以達到減少疫情所帶來的損失之目的。
本論文研究目的在建立一個台灣COVID-19本土日新增確診數的預測模式，利用Unobserved Component Models (UCM，以下簡稱UCM)做為架構，而該預測模式所建立的預測區間，能夠以檢測確診數之實際值是否超出預測區間的方式來當作疫情是否出現異常的依據，用來警示可能發生的疫情爆發。在建立 COVID-19的確診數、死亡數等等預測模式的研究從2020年初至今有許多已發表的文獻，使用的方法包括時間序列、迴歸、機器學習等等，而目前在台灣碩博士論文網上亦或者是國外文獻中尚未有利用UCM來對台灣COVID-19本土日新增確診數做預測模式的相關研究，故進行此研究主題。
本研究使用UCM作為模式架構，該模式包括趨勢、迴歸、週期、季節、隨機誤差項。本研究所使用資料為台灣COVID-19本土日新增確診數，通過觀測台灣日確診數歷史數據得到:日確診數資料具有7天為循環的週期性、受到國定假日與連續假期的影響、受到星期幾的影響、日確診數資料間具有相關性、沒有明顯趨勢、沒有明顯的季節性；並將所觀察到的資料特徵加入UCM架構當中。
基於本研究之UCM架構建立了台灣COVID-19本土日新增確診數的三個適合的配適模式，在經過各項績效指標的比較後選出最佳的預測模式，並對其進行模式驗證，結果顯示，利用本研究所展示之台灣COVID-19本土日新增確診數預測模式所建立的預測區間在30天內皆無實際值超出，顯示出本研究之預測模式的正確性。
本研究所建立之台灣COVID-19本土日新增確診數預測模式能夠在未來COVID-19出現新的疫情時即時的在台灣進行預測，且本研究所提出之預測模式架構能夠在未來爆發類似之疫情時作為參考，提供更具彈性和適應性的預測模式，以因應可能的新變異。

關鍵字

新冠肺炎

並列摘要

The outbreak of COVID-19 since early 2020 has caused various adverse impacts globally, including economic, health, and psychological consequences. Establishing predictive models for the number of confirmed cases and deaths is crucial for early detection and mitigation of outbreaks, aiming to minimize the losses caused by the pandemic.
The purpose of this study is to establish a predictive model for the daily increase in confirmed COVID-19 cases in Taiwan. The model utilizes Unobserved Component Models (UCM) as its framework. The predictive intervals generated by this model serve as indicators of anomaly detection, alerting authorities to potential outbreaks when the actual number of confirmed cases exceeds the predicted range. There are many published documents on establishing prediction models for the number of confirmed cases, deaths, etc. of COVID-19 from the beginning of 2020 to the present. The methods used include time series, regression, machine learning, etc., and currently on the Taiwan Master and Doctoral Thesis Network It may be that there is no relevant research in foreign literature on using UCM to predict the number of daily new confirmed cases of COVID-19 in Taiwan, so this research topic is carried out.
The UCM framework employed in this study includes components for trend, regression, cycle, seasonality, and random error. Daily data on domestic COVID-19 cases in Taiwan were analyzed, revealing cyclic patterns over a 7-day period, impacts from national holidays and consecutive vacation periods, weekday effects, interrelatedness between daily case counts, absence of significant trends, and no apparent seasonality. These observed data characteristics were integrated into the UCM framework.
Based on the UCM framework developed in this study, three well-fitted models for predicting daily increases in confirmed COVID-19 cases in Taiwan were identified. After comparing various performance indicators, the best predictive model was selected and subjected to model validation. The results demonstrate that the predictive intervals established by the model did not exceed the actual values within 30 days, indicating the accuracy of the predictive model proposed in this study.
The predictive model for daily increases in confirmed COVID-19 cases in Taiwan established in this study enables timely forecasting of future outbreaks. Moreover, the proposed model framework can serve as a reference for addressing similar epidemics in the future, providing a more flexible and adaptable predictive model to cope with potential new variants.

並列關鍵字

COVID-19

參考文獻

[1] 魏武雄，2012。時間序列分析：單變量與多變量方法。台北市：智勝文化。

[2] 台灣衛生福利部疾病管制署網站之統計專區，2023。依照發病日所統計之嚴重特殊傳染性肺炎(民國112年3月19號以前病例定義版本)。下載日期：2023年6月6號。來源網址：https：//nidss.cdc.gov.tw/nndss/disease?id=19CoV。

[3] Ameesh, I., Aaqib, M., Dhrubajyoti, B., Akshay, G., Adrija, H., Johanna, C., Gregg, M. L., Gregg, C. F., 2023. A Comparison of COVID-19 and Influenza-Associated Myocarditis: A Nationwide Study in the United States, Current Problems in Cardiology 48(7), 101680.

[4] Abe, C., Shimatani, K., Tsumura, K., Takaguchi, K., Nakayama, Y., Hayashi, T., Mori, C., Suzuki, N., 2024. Impact of COVID-19 on the mental health of primary schoolchildren during the later phase of the pandemic: A case report of an 18-month longitudinal survey in a Japanese primary school, Public Health in Practice 7, 100471.

[5] Akaike, H., 1974. A New Look at the Statistical Model Identification. IEEE Transactions on Automatic Control 19(6), 716-723.