Abstract: Based on the volatility of historical financial indicators, the author constructs a model to identify financial fraud by selecting manufacturing listed companies and their non-fraudulent companies which were punished for the first time from 2007 to 2015 in Shanghai and Shenzhen stock market. The results show that: The historical anomaly data has a good effect on fraud detection; the identification technology does not necessarily improve the efficiency of fraud detection when the anomaly data is improved alone;In order to improve the efficiency of fraud identification, classification technology and data characteristics should be improved to improve their compatibility. In the case of eliminating collinearity, the more advanced the identification technology is, the higher the efficiency of fraud identification will be. The overall recognition rate of financial fraud in the principal component support vector machine model is 85.7%, higher than the recognition effect of improving one aspect alone.

Key Words: Financial Report Fraud; historical abnormal information; classification technology improvement; data information optimization; manufacturing listed company