[Purposes]The paper aims to improve the identification accuracy of black spots in highway traffic accidents. [Methods]Based on highway traffic accident data,abnormal driving behavior data and highway facility data were combined to create a highway black spot feature dataset, and the CatBoost algorithm for ensemble learning was applied to the black spots identification of G4 highway in Hunan Province. [Findings]Compared with the support vector machine, random forest and gradient boosting decision tree, CatBoost has an accuracy of 81% in identifying the black spots of highway accidents, the F₁ value is 0.88,and the A_UC value is increased by 17%, 13% and 17%,respectively, it shows that its identification effect is better than that of the other three machine learning algorithms. [Conclusions] The identification method of black spots in highway traffic accidents based on multi-source data fusion can effectively identify the black spots on the highway,and output the contribution of each accident-related variable to the results. The research results can be applied to screen black spots in highway accidents and guide managers to formulate preventative measures.