我國營所稅稽核之研究

淡江大學機構典藏 > 商管學院 > 資訊管理學系暨研究所 > 學位論文 > Item 987654321/114510

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/114510

题名:	我國營所稅稽核之研究
其它题名:	The study on tax- auditing of profit-seeking enterprise income tax in Taiwan
作者:	李冠諭;Lee, Kuan-Yu
贡献者:	淡江大學資訊管理學系碩士班黃明達;Hwang, Ming-Dar
关键词:	Auditing;Bayesian Network;profit-seeking enterprise income tax;貝氏網路;稽核;營所稅
日期:	2017
上传时间:	2018-08-03 14:54:54 (UTC+8)
摘要:	財政資訊中心提供全國去識別化之報稅資訊給需求單位申請，其中會計師簽證、普通申報等營所稅申報類別皆可能發生逃漏稅行為，但由於稽核人力有限，再加上稽徵機關過往較難產生合理懷疑之抽查理由，因而本研究希望利用資訊科技開發一套稽核模式，並產生抽查名目及重點抽查項目，協助國稅局進行稽核。經評估比較曾應用於營所稅稽核之4種演算法後，決定選用貝氏網路(Bayesian Network)建立稽核預測模型。以全國100年至102年被國稅局抽查到之案件建構全國稽核預測模型；也依行業、縣市分別建立39種分類稽核預測模型，並找出6項能辨別是否需補稅之變數。運用全國103年被國稅局抽查到之案件套入全國稽核預測模型進行驗證後準確度達87.2%。再將102、103年之預估補稅金額逐一加總進行驗證後，誤差率皆低於10%。將全國104年申報案件套入模型依預估補稅金額大小排序挑選出前0.1%之案件，配合分類模型產出精準化抽查名單共531件，預估帶來約117億元稅收，平均每件約2,100萬元，並針對個別案件產出建議重點項目抽查順序供國稅局參考。 Fiscal Information Agency provides the nationwide de-identification tax-related information to units to apply. Enterprise income tax evasion may occur in the declarations of Certified Public Accountants, ordinary declaration, expanded paper review, and others. The purpose of the study was to use information technology to assist National Taxation Bureau under limited resources to do the auditing efficiently by developing auditing models and find a legitimate reason for auditing. This study uses Bayesian Network Algorithm to build the model after comparing to 4 Algorithms applied to tax-auditing of enterprise income tax. The study uses the cases declared between 2011 and 2013 as training data to build the nationwide auditing model and we also find 6 statistically significant attributes. In addition, according to company’s industry and location, 39 classified models were built separately. Then we put the cases declared in 2014 as the testing data into the model to check the accuracy of model, and the accuracy rate is 87.2 %. The predicted sum of tax repayment error ratio of cases in 2013 and 2014 is below 10%. We put the cases declared in 2014 into model, sort the predicted value of repayment in descending order and select top 0.1% as the model precision is close to 90%. Combined with 39 classified models, we could output a more precise checklist of potential tax-evasion companies(531 cases),which is estimated to bring 11.7 billion of tax repayment. (an average of 210 million per case). Besides, according to single cases, we provide key checking items for National Taxation Bureau to prioritize the auditing process.
显示于类别:	[資訊管理學系暨研究所] 學位論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	120	检视/开启

在機構典藏中所有的数据项都受到原著作权保护.

TAIR相关文章

数据加载中.....