Abstract:The security monitoring data of Dangerous chemicals business has great social value, especially real-time accurate prediction of the security risk has become a hot warning research. From the view of four dimensions which are people, equipment, the environment and management, this article analyzes the relevant indicators of safety hazards warning, constructs the bottom-up decision tree based on multi-classification SVM warning model, constructs a bottom-up decision tree SVM multi-classification model based on early warning, to achieve the security level of accurate classification and for future production safety status warning. By comparison with more top-down classification model, it confirms that early warning model used in this paper has better performance in real-time and accuracy, and meets the basic requirements of early warning models.