東方果實蠅一直以來是令果農感到頭痛的害蟲，其危害之後造成的鮮果腐爛，嚴重影響鮮果的經濟價值，過去為預防這樣的情況發生，果農普遍採用噴灑殺蟲劑來防治蟲害，但是這種撲殺的方式，不但會造成藥劑殘留，危害消費者健康，若長期過量使用，也會產生抗藥性，危害農民健康，對於我們生存的環境也會造成嚴重汙染，形成農民、消費者和土地三輸的局面。
過去對於東方果實蠅可能造成影響的氣象因子，如溫度、濕度和雨量，大多採用簡單線性迴歸分析，對於不同空間(農田)及時間(長期重複監測)因素所造成的影響並未加以考慮。有鑑於機器學習逐漸成為資料分析常用的方法，其主要基本概念是讓機器自行學習，從輸入的資料當中找出變數之間的關聯性，並生成貼近現實的資料，所以本研究分別使用線性混合效果模型（樹）、決策樹、隨機森林、極限梯度提升以及深度學習的全連接神經網路等演算法，試圖找出預測最準（誤差最小）及解釋性佳的預測模型。
最後，期望透過本研究建立的模型，對於重要氣象影響因子如溫度和雨量進行分析，提供農民參考並做出相對應的田間管理措施預作防範。

The Bactrocera dorsalis hendel has always been the most troublesome pest for fruit farmers. It causes fresh fruit to rot and seriously affects the economic value of fresh fruit. In the past, to prevent such a situation, many pesticides are sprayed by fruit farmers to control pests, causing pesticide residues, endangering the health of consumers, and producing drug resistance, which results in a three-lose situation for farmers, consumers and land.
In the past, the meteorological factors for Bactrocera dorsalis hendel that may have an impact, such as temperature, humidity, and rainfall, were mostly analyzed by simple regression, without considering different spaces (farmland) and time (long-term repeatability). Given that machine learning has become a common method for data analysis, its main concept is to let the machine learn by itself and find the correlation between variables from input data, and then generate data that match reality. Therefore, in this study, the linear mixed-effects model tree, decision tree, random forest, XGBoost and fully connected neural network are used to build models for prediction and interpretation.
Finally, through the models proposed in this study, it is expected that farmers can make corresponding field management strategies in their farmland through the most important meteorological factors such as temperature and rainfall.

論文審定書 i
誌謝 ii
摘要 iii
Abstract iv
目錄 v
圖次 vii
表次 viii
第一章 緒 論 1
1.1 研究背景 1
1.2 研究動機 2
1.3 研究目的 2
第二章 文獻探討 3
2.1 相依資料與時間序列 3
2.2多層次嵌套資料 4
2.3 機器學習演算法 6
2.4 交叉驗證(Cross validation) 8
2.5 Prequential 9
第三章 研究方法與步驟 10
3.1 研究流程 11
3.2 研究方法 11
3.3 評估模型標準 13
第四章 研究結果與分析 14
4.1資料蒐集 14
4.2資料清理 14
4.3實驗設計 14
4.4 建立模型 16
4.5 評估模型 18
4.6模型解釋 19
4.7 Ensemble Model Selection 23
第五章 討論與建議 24
5.1 研究結論 24
5.2 未來建議 25
5.3 研究限制 27
參考文獻 27

[1]行政院農業委員會農業試驗所. 東方果實蠅區域整合性蟲害管理策略[EB/OL]. 行政院農業委員會農業試驗所. 行政院農業委員會農業試驗所2014-02-13/2022-06-13. https://www.tari.gov.tw/form/index-1.asp?Parser=2,6,1113,1106,,,4061,,,,2.
[2]東方果實蠅生態及為害特性簡介-Chapter 1.pdf[J].
[3]東方果實蠅整合性管理與防治策略 (2013).pdf[J].
[4]東方果實蠅防治作業手冊.pdf[J].
[5]鄭允、黃毓斌、高靜華、蘇文瀛,東方果實蠅防治作業手冊,2000
[6]陳正昌、林曉芳(2020) R統計軟體與多變量分析,初版
[7]邱皓政(2021) 多層次模式與縱貫資料分析:Mplus 8解析應用, 初版
[8]吳明隆、張毓仁(2014) 多層次模式的進階應用, 初版
[9]謝俊義((2015)多層次分析-理論、方法與實務, 二版
[10]江崎貴裕(2021)資料科學的建模基礎-別急著coding，你知道模型的陷阱嗎? 初版
[11]Hiromu Nishiuchi(2021) 機器學習的數學基礎,初版
[12]何宗武(2017) R資料採礦與數據分析, 二版
[13]何宗武(2018) 大數據決策分析盲點大突破10講,初版
[14]應用寄生性天敵昆蟲防治東方果實蠅之策略(農委會)，從https://www.coa.gov.tw/ws.php?id=23736
[15]Mittler R. Abiotic stress, the field environment and stress combination[J]. Trends in Plant Science, 2006, 11(1): 15–19.
[16]Fand B B, Kamble A L, Kumar M. Will climate change pose serious threat to crop pest management: A critical review?[J]. 2012, 2(11): 16.
[17]Baldacchino, F., Krčmar, S., Bernard, C., Manon, S., & Jay-Robert, P. (2017). The impact of land use and climate on tabanid assemblages in Europe. Agriculture, Ecosystems & Environment, 239, 112–118
[18]Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67, 1–48.
[19]Capitaine, L., Genuer, R., & Thiébaut, R. (2021). Random forests for high-dimensional longitudinal data. Statistical Methods in Medical Research, 30(1), 166–184.
[20]Cerqueira, V., Torgo, L., & Mozetic, I. (2020). Evaluating time series forecasting models: An empirical study on performance estimation methods. Machine Learning, 109(11), 1997–2028.
[21]Correlated Data Analysis: Modeling, Analytics, and Applications. (2007). Springer New York.
[22]Fitting GLMM trees using R package glmertree.
[23]Gu, S., Han, P., Ye, Z., Perkins, L. E., Li, J., Wang, H., Zalucki, M. P., & Lu, Z. (2018). Climate change favours a destructive agricultural pest in temperate regions: Late spring cold matters. Journal of Pest Science, 91(4), 1191–1198.
[24]Hidalgo, B., & Goodman, M. (2013). Multivariate or Multivariable Regression? American Journal of Public Health, 103(1), 39–40
[25]Makumbe, L. D. M., Moropa, T. P., Manrakhan, A., & Weldon, C. W. (2020). Effect of sex, age and morphological traits on tethered flight of Bactrocera dorsalis (Hendel) (Diptera: Tephritidae) at different temperatures. Physiological Entomology, 45(2–3), 110–119.
[26]Miao, J., Huang, J., Wu, Y., Gong, Z., Li, H., Zhang, G., Duan, Y., Li, T., & Jiang, Y. (2019). Climate factors associated with the population dynamics of Sitodiplosis mosellana (Diptera: Cecidomyiidae) in central China. Scientific Reports, 9(1), 12361.
[27]Philipp, M., Zeileis, A., & Strobl, C. (2016). A toolkit for stability assessment of tree-based learners. 11.
[28]REENA, Kaur, A., Singh, M., Sinha, B. K., Kumar, A., & Ahmad, S. (2020). Impact of abiotic factors on population dynamics of Bactrocera dorsalis Hendel and Bactrocera zonata (Saunders) at different ecological zones in NW Plains of India. Journal of Agrometeorology, 22(3), 250–257.
[29]Sagi, O., & Rokach, L. (2018). Ensemble learning: A survey. WIREs Data Mining and Knowledge Discovery, 8(4), e1249.
[30]Samayoa, A. C., Choi, K. S., Wang, Y.-S., Hwang, S.-Y., Huang, Y.-B., & Ahn, J. J. (2018). Thermal effects on the development of Bactrocera dorsalis (Hendel) (Diptera: Tephritidae) and model validation in Taiwan. Phytoparasitica, 46(3), 365–376.
[31]Snijders, T., & Bosker, R. (1999). Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. http://lst-iiep.iiep-unesco.org/cgi-bin/wwwi32.exe/[in=epidoc1.in]/?t2000=013777/(100).
[32]Winter, B. (2013). Linear models and linear mixed effects models in R with linguistic applications. ArXiv:1308.5499 [Cs].
[33]Xu, Y., Zafirov, A., Alvarez, R. M., Kojis, D., Tan, M., & Ramirez, C. M. (2020). FREEtree: A Tree-based Approach for High Dimensional Longitudinal Data With Correlated Features. arXiv:2006.09693 [cs, stat].
[34]Fokkema, M., Smits, N., Zeileis, A., Hothorn, T., & Kelderman, H. (2018). Detecting treatment-subgroup interactions in clustered data with generalized linear mixed-effects model trees. Behavior Research Methods, 50(5), 2016–2034.
[35]lmerControl function—RDocumentation.
[36]James L. Peugh(2010) A practical guide to multilevel modeling
[37]Introduction to Linear Mixed Models[EB/OL]. /2022-07-06. https://stats.oarc.ucla.edu/other/mult-pkg/introduction-to-linear-mixed-models/.
[38]https://ourcodingclub.github.io/tutorials/mixed-models/#what
[39]https://www.baphiq.gov.tw/publish/plant_protect_pic_9/orangePDF/02-16.pdf
[40]https://ithelp.ithome.com.tw/articles/10273094
[41]https://www.tinytsunami.info/fully-connected-neural-network/
[42]https://blog.csdn.net/weixin_44751294/article/details/125086398