運用決策樹於營利事業所得稅結算申報書選案查核之研究

• Main Authors: Pei-Fen Liu, 劉珮芬
• Other Authors: Kuo-Tay Chen
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/75946442672572229975

碩士 === 國立臺灣大學 === 會計學研究所 === 91 === Business income tax is the main tax revenue for government. In order to promote the efficiency of tax administration, the government has established the case selection system to examine the business income tax returns. The purpose of this research is to study the classification ability of decision tree models and derive the behavior of tax evasion through classification rules. Using the firm-level data from the 1996 business income tax returns compiled by the Public Finance and Tax Data Processing Center, Ministry of Finance, and the data mining software ”Answer Tree 2.1” to establish the Exhaustive CHAID models, this study empirically demonstrates that decision tree is a practical tool for tax audit. By using the classification rules to select 10% cases for audit, we could identify half of the tax evasion dollar amount. The finding also indicates that industry categories, entertaining expenses/total sales ratio and filing styles are the factors that can significantly discriminate the behavior of tax evasion. Based on the empirical findings, the government can improve the effectiveness of case selection system and secure the tax revenue by applying the decision tree models to selection of business tax returns for audit.