After occurrence of natural disasters,active distribution network (ADN) can promptly restore power supply to some critical loads through tie-line switching and flexible distributed generation (DG),and thus the fault risk is effectively mitigated. A data-driven multi-dimensional intelligent forecast approach for the fault risk levels in ADNs is proposed in this paper. Firstly,a feature selection method based on Chi-square test (χ²) and Pearson correlation coeffects is developed to analyze the strength of fault correlation factors from multiple dimensions and the optimal set of fault features is obtained. Then,an optimal network reconfiguration model is established for the damaged ADNs considering DG integration,and consequently the heterogeneity of the line importance can be taken into account which provides a solid foundation for the classification of fault risks. Furthermore,an intelligent forecast model for ADN fault risk levels is established based on extreme gradient boostig (XGBoost) algorithm. Finally,the numerical tests on IEEE RBTS Bus6 distribution network demonstrate that the proposed approach achieves a predication accuracy 3.17% higher than back propagation (BP) neural network does. The proposed approach has good generalization capability,thus providing an important basis for the fault risk management in ADNs to effectively reduce the fault loss.