Comprehensive risk assessment of natural disasters based on machine learning in Changzhou City, China

The ecological environment is continuously deteriorating, leading to a rise in the frequency of natural disasters in recent years. In this study, five different machine learning regression models (random forest, extra trees, support vector regression, XGBoost, and LightGBM) are used to assess the risk of four types of natural disasters in Changzhou. Utilizing ArcGIS software, a risk level map for each individual disaster is generated. Moreover, based on consulting local experts, the comprehensive risk of natural disasters in Changzhou is evaluated by the weighting method. Finally, the accuracy of the model is verified by historical disaster point data, and the accuracy is judged by the area under curve (AUC). The results show that the accuracy of the XGBoost regression models is the highest, and the values of R² are 0.940, 0.873, 0.928, and 0.893, respectively. In the comprehensive risk assessment of natural disasters in Changzhou City, Nandu Town, Xuebu Town, and Hengshanqiao Town are identified as the regions with the highest risk areas, encompassing 7361.64ha, 5551.02ha, and 5152.59ha, respectively. The AUC value under the comprehensive risk level is 0.723, which has good accuracy and can be used to evaluate the comprehensive risk in Changzhou. This result provides a reference and guidance for the local government.