Abstract: Although the transient angle stability assessment of power systems based on data-driven can provide more accurate results, it is difficult to apply in engineering practice due to the lack of interpretability of the results. To solve this problem, an analysis method of transient angle stability assessment and interpretability based on gradient boosting enhanced with step-wise feature augmentation (abbr. AugBoost) was proposed. Firstly, the mapping relationship between the power system input features and transient angle stability index was established by training the AugBoost assessment model. Secondly, the real-time measurement data of phasor measurement units were transferred to the trained AugBoost assessment model to provide real-time assessment results. And the relationship between assessment results and input features was explained according to Shapley additive explanations (abbr. SHAP) interpretation model to improve the credibility of results. Finally, a model update process was designed to improve the robustness of the assessment model to the variation of power system operating conditions. The simulation results on the 23-bus power system and the 1648-bus power system provided by power system simulation software PSS/E verify the effectiveness of the proposed method.