With the enhancement of social environmental protection consciousness, electric vehicles using clean energy are promoted widely. Scope of time-sharing leasing business for the electric vehicle keeps expanding, and multi-level management platforms are set up one after another. Electric vehicle parameters collected by main management platform are accumulated continually, which lays a firm foundation for big data analysis research. Preprocessing technology plays an important role in the process of data acquisition. Specially, monitoring data collected from the each sub-platform of operators may have the risk of being tampered manually. Therefore, in this paper, human intervention probability curve is used to quantify the relationship between monitoring data and covariance ratio, and both of human intervention probability curve interval and observation data relationship are treated as input, so that the regression object decision tree is established and introduced to the traditional Kalman filter algorithm. Therefore, a modified Kalman filter prediction method based on decision tree analysis is presented which can be used to reduce the influence of human tampering, and achieve preprocessing function of main operating platform for collecting data. In this paper, this modified method is successfully used in the field of vehicle speed prediction, and used to get speed data with higher credibility, which provides a strong technical support for large data analysis.