流感及其併發症病例數之研究與預測

Study and Forecast of Cases of Flu and Their Complications

鄒佳蓁(Jia-Zhen Zou) ； 柯沛程(Jau-Chuan Ke) ； 簡郁紘(Yu-Hung Chien) ； 湯涴貽(Wo-Yi Tang) ； 陳怡如(Yi-Ru Chen) ； 曾瑋慈(Wei-Cih Zeng) ； 陳宛晴(Wan-Cing Chen)

《中國統計學報》 59卷3期 (2021/09) Pp. 172-225

摘要

本研究使用Python程式語言進行時間序列分析，以四種全國流感及其併發症病例數爲研究對象，分別爲門診人數、急診人數、併發重症本土病例數、死亡人數，資料來源爲中華民國衛福部疾管署傳染病統計資料查詢系統，樣本期間選自2008／1週至2020／28週；預測期間則自2020／29週至2020／33週。研究發現門診人數、急診人數、死亡人數皆具有季節性，併發重症本土病例數具間歇性，彼此間之變化具關聯性，且雖然死亡人數呈現逐年攀升的趨勢，但觀察其時間序列圖形的波動，卻可發現其波動幅度與前述二者相似且呈現遞减的趨勢。本研究共使用三種預測模型，包含ARIMA模型，Croston模型、Holt-Winters' seasonal模型，並針對預測期間之真實值與模型預測值做比較，以評估模型預測能力。在ARIMA模型的分析結果中，發現門診人數模型在三者時間序列預測模型中具較佳的預測能力。在Croston模型與Holt-Winters' seasonal模型的分析結果比較中，發現Holt-Winters' seasonal模型具較佳的預測能力。

關鍵字

時間序列 ； 流感及其併發症 ； 預測能力 ； ARIMA模型 ； Croston模型 ； Damped ； Holt-Winters' seasonal模型 ； Python ； Syntetos-Boylan approximation（SBA）

並列摘要

Python is used to analyze time series in this study. This study mainly looks into the number of people who suffered from flu and their complications in Taiwan; The patients were categorized into four groups: Outpatient, Emergency, Severe Complicated Influenza, and Death. The source of data is from the Taiwan National Infectious Disease Statistics System of Centers for Disease Control, Ministry of Health and Welfare of Republic of China. The sample in this study has been collected from 1st week, 2008 to 28th week, 2020, and the forecast period was from 29th week, 2020 to 33rd week, 2020. This study found that the Outpatient, the Emergency, and the Death have changed seasonally, while the Severe Complicated Influenza has changed intermittently. All of their changes are correlated. In addition, although the toll has clinged throughout the years, if we observe their graphs of time series, we can find that the fluctuations of death are similar to those of Outpatient and Emergency, and that they have all been decreasing. Three prediction models were used in this study, including ARIMA model, Croston model and Holt-winters' seasonal model. This study compares real value and prediction model's value in the forecast period, and estimates the prediction ability of the models. The analysis results of the ARIMA model, show that Outpatient people model have the best prediction ability. In addition, Croston model and Holt-winters' seasonal model, were compared and the results suggest that the prediction ability of the latter is better than the former.

並列關鍵字

ARIMA model ； Croston model ； Damped ； Flu and its complications ； Holt-Winters' seasonal model ； Prediction ability ； Python ； Syntetos-Boulanger approximation (SBA) ； Time series

參考文獻

Waller, D. (2015). Methods for Intermittent Demand Forecasting. URL: https://www.lancaster.ac.uk/pg/waller/pdfs/Intermittent_Demand_Forecasting.pdf

Winters, P. R. (1960). Forecasting sales by exponentially weighted moving averages. Management Science, 6, pages 324–342.

World Health Organization.(2018, November 6).Influenza (Seasonal).URL: https://www.who.int/en/news-room/fact-sheets/detail/influenza-(seasonal)

吳齊軒 (2010)。 類流感高峰預測之相關議題。

社團法人台灣感染管制學會 (2017)。 老年族群之傳染病研究。 取自 https://www.cdc.gov.tw/Professional/ProgramResultInfo/LeYn5b0UwF_lgvjR5rhT-A?programResultId=YLbKJe5cMfaa5nl6fBPTIQ