臺灣博碩士論文加值系統

利用台灣上市、櫃公司財務報表重編作為審計失敗的指標，並依照Fully Connected Feedforward Network架構架設深度學習模型，用以預測可能發生審計失敗的查核案件，並利用半監督式學習與Voting等方式強化預測效果。與對照組邏輯斯回歸模型相比，預測能力顯著提升。

Used financial statement restatements of Taiwanese lised companies as indicator of audit failure, and built a Deep learning models based on Fully Connected Feedforward Network framework to predict audit failure, then used semi-supervised learning and Noting methods to improve prediction outcome. The predictive ability was signigicantly improved compared with the logistic regression model of the control group.

口試委員會審定書 I
中文摘要 III
英文摘要 IV
感謝詞 V
圖目錄 IX
表目錄 X
第一章 緒論 1
第一節 研究背景與動機 1
第二節 研究目的 2
第三節 研究架構 3
第二章 文獻回顧 4
第一節 審計失敗 4
第二節 審計失敗的傳染效果 5
第三節 裁決性應計數 6
一、 Johns Model 6
二、 Modified Johns Model 7
三、 加入ROA的Modified Johns Model 7
第三章 深度學習 8
第一節 名稱的由來 8
第二節 Activation Function 9
第三節 深度學習架構 10
一、 Input layer 11
二、 Hidden layer 11
三、 Output layer 11
四、 Fully Connected Feedforward Network 12
第四節 Loss Function 13
第五節 Gradient Descent 14
第六節 Backpropagation 15
第七節 小結 17
第四章 研究方法 18
第一節 資料整理 18
一、 審計失敗 18
二、 裁決性應計數 19
三、 審計失敗傳染效果 20
四、 整理結果 21
第二節 模型架構 22
第三節 訓練模型 24
一、 資料清洗 25
二、 訓練資料抽選 25
三、 半監督式學習 26
四、 訓練過程 28
五、 Voting 30
第五章 研究結果 32
第一節 實驗組 32
第二節 對照組 32
第三節 實驗組與對照組比較 35
第四節 模組的優點與缺點 37
第六章 結論 38
第一節 研究結論 38
第二節 研究建議 38
參考資料 39
附錄 41
Confusion Matrix 41
Zmijewski (1984) index 42
其他參數 43

Bishop, C. M. 2006. Pattern recognition and machine learning: springer.
bravotty. 2020. Information-entropy-loss-pytorch, March 31 2020 [cited December 15 2020]. Available from https://github.com/bravotty/Information-entropy-loss-pytorch/blob/master/entropy_loss_pytorch.py.
De Fuentes, C., and R. Porcuna. 2019. Predicting audit failure: evidence from auditing enforcement releases. Spanish Journal of Finance and Accounting/Revista Española de Financiación y Contabilidad 48 (3):274-305.
DeAngelo, L. E. 1981. Auditor size and audit quality. Journal of accounting and economics 3 (3):183-199.
Dechow, P. M., R. G. Sloan, and A. P. Sweeney. 1995. Detecting earnings management. Accounting review:193-225.
Glorot, X., and Y. Bengio. 2010. Understanding the difficulty of training deep feedforward neural networks. Paper read at Proceedings of the thirteenth international conference on artificial intelligence and statistics.
Hinton, G. E., S. Osindero, and Y.-W. Teh. 2006. A fast learning algorithm for deep belief nets. Neural computation 18 (7):1527-1554.
Hoskiss. 2020. [機器學習] Backpropagation with Softmax / Cross Entropy 2019 [cited December 2 2020]. Available from https://medium.com/hoskiss-stand/backpropagation-with-softmax-cross-entropy-d60983b7b245.
Jones, J. J. 1991. Earnings management during import relief investigations. Journal of accounting research 29 (2):193-228.
Kothari, S. P., A. J. Leone, and C. E. Wasley. 2005. Performance matched discretionary accrual measures. Journal of accounting and economics 39 (1):163-197.
Krishnan, G. V. 2003. Audit quality and the pricing of discretionary accruals. Auditing: A journal of practice & theory 22 (1):109-126.
Li, L., B. Qi, G. Tian, and G. Zhang. 2015. The Contagion Effect of Low-Quality Audits along Individual Auditors. Available at SSRN 2478348.
ML-Glossary. Loss Functions 2017 [cited. Available from https://ml-cheatsheet.readthedocs.io/en/latest/loss_functions.html.
Nielsen, M. CHAPTER 4 A visual proof that neural nets can compute any function 2019 [cited. Available from http://neuralnetworksanddeeplearning.com/chap4.
Rumelhart, D. E., G. E. Hinton, and R. J. Williams. 1986. Learning representations by back-propagating errors. nature 323 (6088):533-536.
Saito, T., and M. Rehmsmeier. 2015. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PloS one 10 (3):e0118432.
ufoym. 2020. Imbalanced Dataset Sampler, October 9 2020 [cited December 15 2020]. Available from https://github.com/ufoym/imbalanced-dataset-sampler.
Zmijewski, M. E. 1984. Methodological issues related to the estimation of financial distress prediction models. Journal of accounting research:59-82.
吳琮璠教授. 2001. 審計學--新觀念與本土化. 台北市: 吳琮璠教授.
李宏毅. 2016a. ML Lecture 1: Regression - Case Study. YouTube. https://www.youtube.com/watch?v=fegAeph9UaA.
———. 2016b. ML Lecture 2: Where does the error come from? . YouTube. https://www.youtube.com/watch?v=D_S6y0Jm6dQ.
———. 2016c. ML Lecture 3-1: Gradient Descent. YouTube. https://www.youtube.com/watch?v=yKKNr-QKz2Q.
———. 2016d. ML Lecture 6: Brief Introduction of Deep Learning. YouTube. https://www.youtube.com/watch?v=Dr-WRlEFefw.
———. 2016e. ML Lecture 7: Backpropagation. YouTube. https://www.youtube.com/watch?v=ibJpTrp5mcE.
———. 2016f. ML Lecture 9-1: Tips for Training DNN. YouTube.
———. 2016g. ML Lecture 11: Why Deep? YouTube. https://www.youtube.com/watch?v=XsC9byQkUH8.
———. 2016h. ML Lecture 12: Semi-supervised. YouTube. https://www.youtube.com/watch?v=fX_guE7JNnY.
———. 2017a. ML Lecture 5: Logistic Regression. YouTube. https://www.youtube.com/watch?v=hSXFuypLukA.
———. 2017b. ML Lecture 22: Ensemble. YouTube. https://www.youtube.com/watch?v=tH9FH1DH5n0.
陳怡均. 2008. 簡介美國PCAOB 對於公開公司會計師之監理. 金融監督管理委員會 2008 [cited October 16 2008]. Available from https://www.fsc.gov.tw/fckdowndoc?file=/%E5%AF%A6%E5%8B%99%E6%96%B0%E7%9F%A5%20(1).pdf&flag=doc.
蔡孟瑾. 2015. 審計失敗之傳染效果-以台灣為例. 臺灣大學會計學研究所學位論文:1-48.