Abstract: Risk factors leading to operational vehicle accidents come from many aspects, such as people, vehicles, environment, etc. Traditional traffic accident prediction models cannot intuitively depict the coupling relationship between various risk factors. Therefore, this paper studies the coupling degree model based on Random Forest (RF). Quantify the coupling effects of multiple risk factors and their effects on accident severity. Firstly, the data of 5186 operational vehicle accidents in Virginia from 2021 to 2023 were preprocessed, and 18 types of accident risk first-level index characteristics were selected from three aspects: human, vehicle and environment. RF model was used for feature screening, and the optimal feature set was established. Then, with the optimal selection as the independent variable and the accident severity as the dependent variable, an RF prediction model is constructed to predict the accident severity of operational vehicles, and the correlation between risk factors and accident severity is studied. The relevant characteristics were sorted according to importance, and the 15 most important risk factors were selected according to class to conduct coupling relationship research, and the impact of coupling effect between different risk factors on accident severity was quantified. The results show that RF prediction model is superior to GBDT and SVM models in predicting accident severity. Secondly, in the coupling relationship, when there are six types of factors, such as "serious speeding", "fatigue driving", "failure to maintain a safe distance", "uphill and downhill turns", "road ice" and "not wearing seat belts", the severity of the accident is more likely to be aggravated, among which "serious speeding" is most likely to form a strong coupling with other factors, resulting in serious injury accidents.