Aiming at the current problems of frequent traffic accidents, high subjectivity of risk warning features and low accuracy of road grade prediction, a random forest-matter-element risk warning model is proposed. The model selects the actual data set of the I880-N highway in the United States, defines 13 warning evaluation indicators from three aspects: traffic flow, speed and occupancy, and uses the random forest method to screen the warning features to achieve the purpose of removing dimensional disasters and weight distribution. The K-means method is introduced into the matter-element model to divide the risk level. The established model is applied to the accident-prone locations on the main line of the I880-N highway in the United States. The results show that the comprehensive risk goodness of the highway is 0.053, the warning level is level 4, and it has severe risk. This evaluation result is consistent with the actual situation, which verifies the applicability of the risk warning model and can provide new ideas for the development of road risk warning work.