流程效率改善一直是組織管理優化的重點，雖然隨著企業組織數位化轉型，企業使用企業資源軟體(ERP)、商業流程管理軟體(BPM)進行管理，產生大量數位化足跡－系統事件日誌資料(Log Data)，但大部分都使用系統資料庫內的資料作為分析重點，卻很少直接從系統日誌資料討論流程時間改善，此外，組織並非都有明確之標準流程，或是需要由非流程參與者部門人員進行流程效率改善。
本研究以埃因霍溫科技大學(Eindhoven University of Technology)國際差旅費申報事件日誌資料，實證僅提供事件日誌資料，運用目標-問題-特徵表(Goal-Question-Feature Index, GQFI)進行流程診斷，可直接了解流程執行情況，並發現流程瓶頸，另外，為了進行流程效率改善參考，試圖從事件日誌資料中預測流程剩餘時間，考量事件日誌資料為流程活動組合的序列，具有長距離依賴問題，且因為不同流程具有不同活動語義關聯性，使用了ProcessTransformer模型來預測流程的剩餘時間，儘管只使用了二項變數－活動序列和活動已使用時間進行預測，但模型的雙曲餘弦對數(log-cosh)平均值仍然低於0.02，且可隨活動進度進行預測；最後，為了由事件日誌資料中，找出影響流程使用時間的關鍵因素，從案例軌跡、活動事件方向分析，運用邏輯迴歸、分類樹、隨機森林三種分類機率模型，並通過分類樹IF-THEN規則及隨機森林變數重要性，分析出5個關鍵因素，並從其中提出流程改善建議方向；運用本文中三種研究方法，可達到循環式流程效率改善。

Process efficiency improvement has always been the focus of organizational management optimization. Although with the digital transformation of business organizations, enterprises use enterprise resource software (ERP) and business process management software (BPM) for management, generating a large number of digital footprints - system event log data (Log Data), most of them use the data in the system database as the focus of analysis, but rarely discuss process time improvement directly from the system log data. In addition, organizations do not always have clear standard processes or require process efficiency improvement by non-process participant departmental staff.
This study uses data from the international travel reimbursement event log dat at Eindhoven University of Technology to demonstrate that the process diagnosis using the Goal-Question-Feature Index (GQFI) can directly understand the process implementation and identify process bottlenecks by providing only event log data. In addition, in order to improve the process efficiency, we tried to predict the remaining time of the process from the event log data, considering that the event log data is a sequence of process activity combinations, which has the problem of long-range dependencies, and because different processes have different activity semantic associations, we used the ProcessTransformer model to predict the remaining time of the process,. Although only two variables which are the sequence of activities and the used time of activities used to predict , the mean value of log-cosh of the model is still lower than 0.02, and the prediction can be performed with the progress of activities. Finally, in order to find out the key factors affecting the process usage time from the event log data, we analyzed the case trajectory and activity event direction, used three types of classification probability models: logical regression, classification tree, and random forest, and analyzed five key factors through the classification tree IF-THEN rule and the importance of random forest variables, and suggested directions for process improvement from them. By applying the three methods in this study, cyclical process efficiency improvement can be achieved.

論文審定書 i
致謝辭 ii
摘要 iii
Abstract iv
目錄 vi
圖目錄 vii
表目錄 viii
第一章 緒論 1
1.1 研究背景 1
1.2 研究動機與目的 1
第二章 文獻探討 3
2.1 流程探勘 3
2.2 機器學習 5
2.2.1 邏輯迴歸 5
2.2.2 分類樹 5
2.2.3 隨機森林 6
2.3 深度學習－ProcessTransformer模型 6
第三章 研究方法 9
3.1 研究流程及方法 9
3.2 分類演算法模型評估指標 12
第四章 研究結果與討論 14
4.1 資料說明 14
4.2 GQFI流程診斷 16
4.3 流程剩餘時間預測 19
4.4 影響流程使用時間原因探討 23
第五章 結論 32
參考文獻 34

1. Berti, A., van Zelst, S. J., & van der Aalst, W. (2019). Process Mining for Python (PM4Py): Bridging the Gap Between Process- and Data Science (arXiv:1905.06169). arXiv. http://arxiv.org/abs/1905.06169
2. Bukhsh, Z. A., Saeed, A., & Dijkman, R. M. (2021). ProcessTransformer: Predictive Business Process Monitoring with Transformer Network (arXiv:2104.00721). arXiv. http://arxiv.org/abs/2104.00721
3. Erdogan, T. G., & Tarhan, A. K. (2022). Multi-perspective process mining for emergency process. Health Informatics Journal, 28(1), 146045822210771. https://doi.org/10.1177/14604582221077195
4. Hochreiter, S., & Schmidhuber, J. (1997). Long Short-Term Memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
5. Jallow, A. K., Majeed, B., Vergidis, K., Tiwari, A., & Roy, R. (2007). Operational risk analysis in business processes. BT Technology Journal, 25(1), 168–177. https://doi.org/10.1007/s10550-007-0018-4
6. Molnar, C. (2020). Interpretable machine learning. Lulu. com.
7. Paperno, D., Kruszewski, G., Lazaridou, A., Pham, Q. N., Bernardi, R., Pezzelle, S., Baroni, M., Boleda, G., & Fernández, R. (2016). The LAMBADA dataset: Word prediction requiring a broad discourse context (arXiv:1606.06031). arXiv. http://arxiv.org/abs/1606.06031
8. R Core Team. (2022). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. https://www.R-project.org/
9. Reinkemeyer, L. (編輯). (2020). Process Mining in Action: Principles, Use Cases and Outlook. Springer International Publishing. https://doi.org/10.1007/978-3-030-40172-6
10. Sperandei, S. (2014). Understanding logistic regression analysis. Biochemia Medica, 12–18. https://doi.org/10.11613/BM.2014.003
11. Van Der Aalst, W. (2012). Process Mining: Overview and Opportunities. ACM Transactions on Management Information Systems, 3(2), 1–17. https://doi.org/10.1145/2229156.2229157
12. van der Aalst, W. (2016). Process Mining. Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-662-49851-4
13. Van Der Aalst, W., Weijters, T., & Maruster, L. (2004). Workflow mining: Discovering process models from event logs. IEEE Transactions on Knowledge and Data Engineering, 16(9), 1128–1142. https://doi.org/10.1109/TKDE.2004.47
14. Zdravkovic, J., Kirikova, M., & Johannesson, P. (編輯). (2015). Advanced Information Systems Engineering: 27th International Conference, CAiSE 2015, Stockholm, Sweden, June 8-12, 2015, Proceedings (卷 9097). Springer International Publishing. https://doi.org/10.1007/978-3-319-19069-3