Traditional road safety prediction methods have low recall rate and poor prediction accuracy. This paper proposes a road traffic safety risk prediction method based on improved random forest. First, collect road traffic data, such as static data, traffic dynamic data, other traffic related data and accident data. Then, the abnormal road traffic data are identified based on chaos method, and the abnormal data are repaired with grey GM (1, n) model. Finally, the random forest algorithm is improved by optimizing the similarity measurement method a...