摘要： 税收申报欺诈检测是税务机关税收征管和稽查中面临的一个重要问题。该文提出了一种基于SVM 的税收申报欺诈检测方法。首先用采样来的企业经营和财务数据训练好一个SVM 识别系统，然后用这个SVM 识别系统判断一个企业的报税额是否真实。实验结果说明该方法是一种有效的方法，在31 个测试样本中，检测精度达87.10%，比基于See 5.0 的方法高3.23%，而训练时间只需1.708s。

关键词: 模式识别；SVM；欺诈检测；税

Abstract: Fraud detection in tax declaration is an interesting problem. An approach of fraud detection in tax declaration based on a support vector machine (SVM) is proposed in this paper. ASVM is trained using financial data of sampled enterprises, the SVM is then employed to detect whether tax data declared by an enterprise is true or not. Experimental results show that the proposed approach is effective: classification precision of proposed method is 87.10% in 31 sample data, and it is 3.23% higher than that of See 5.0. Training the SVM from 30 sample training data set takes only 1.708s.

Key words: Pattern recognition; SVM; Fraud detection; Tax