为了提升地质灾害易发性评价的精度，以浙江省杭州市富阳区为研究区，提出了考虑缓冲区优化策略的随机森林地质灾害易发性评价方法。首先，选取归一化植被指数、距道路距离、距断层距离、汛期降雨量、坡度、坡向、起伏度、距水系距离和岩性共9个评价因子，通过多重共线性分析确保因子的独立性。其次，构建0.5、1、1.5、2 km 4种缓冲区范围，以随机采样方法生成负样本点，避免正负样本交叉污染，提高样本代表性和模型的区分能力，并增设一组无缓冲区的随机采样点作为参照。接着，采用随机森林算法对地质灾害易发性模型进行训练和测试，结果表明缓冲区优化策略能够显著提升模型预测精度，且缓冲区的设定存在最优边界。其中，1 km缓冲区对应的模型AUC（area under curve）值最高为0.815，表明该缓冲区设定下采集的负样本能够更加准确地区分地质灾害特征。最后，基于最优缓冲区与随机森林算法（random forest, RF）模型的易发性评价结果，高易发区主要集中在西北部和东南部的山区地带，且频率比随着易发性等级升高而不断增大，验证了这一方法的科学性，可为富阳区的地灾防治工作提供依据。

In order to improve the accuracy of geological hazard susceptibility assessment, Fuyang District in Hangzhou, Zhejiang Province was taken as the research area and a random forest method was proposed for evaluating geological hazard susceptibility, considering buffer zone optimization strategies. Firstly, nine evaluation factors were selected: normalized difference vegetation index, distance to roads, distance to faults, rainfall during the flood season, slope, aspect, ruggedness, distance to water systems, and lithology. Multicollinearity analysis was conducted to ensure the independence of the factors. Secondly, buffer zones of 0.5 km, 1 km, 1.5 km, and 2 km were constructed. Negative sample points were generated using random sampling to avoid cross-contamination between positive and negative samples, enhance sample representativeness, and improve the model's discrimination capability. An additional set of random sampling points without buffer zones was also established for comparison. The random forest algorithm was then used to train and test the geological hazard susceptibility model. Results indicated that the buffer zone optimization strategy significantly improved the model's predictive accuracy and that there was an optimal boundary for the buffer zone. The model's AUC(area under curve) value was highest at 0.815 for the 1 km buffer zone, indicating that negative samples collected within this buffer zone could more accurately distinguish geological hazard characteristics. Finally, based on the susceptibility evaluation results of the optimal buffer zone and the random forest model, high susceptibility areas were mainly concentrated in the mountainous regions in the northwest and southeast. The frequency ratio increased with the susceptibility level, validating the scientific validity of this method. This approach can provide a basis for geological hazard prevention and control in Fuyang District.