Journal of Digital Information Management Chen Yao, Jiuchun Gu 13 5 2015 http://dline.info/fpaper/jdim/v13i5/v13i5_10.pdf The real traffic databases are highly susceptible to noisy, missing, and inconsistent data. As reason of processing the discrepant data, a model is built based on square support vector machine which is good at dissolving the problems such as small samples, nonlinear and pattern recognition. Utilizing of the SVM, the inaccurate data can be detected by calculating the difference value between the real and prediction data. Comparing with the method of threshold theory, the model is proved to be better for the accurate data detection online and database cleaning.