In this research, landslide susceptibility map of the Fahliyan sub-basin was provided employing adaptive neuro-fuzzy inference system (ANFIS) in ensemble with the ant colony optimization (ACOR) and differential evolution (DE) algorithms. Forty-three out of 61 landslides (70%) were employed to provide landslide susceptibility map and 18 landslides (30%) to validate the models. Thirteen landslide controlling factors including altitude, plan curvature, slope angle, aspect, profile curvature, distance to roads, distance to rivers, distance to faults, rainfall, TWI, SPI, land use, and lithology were employed to provide the map of landslide susceptibility. Weights of every effective factor class and effective factors were calculated based on frequency ratio of landslides relative to the class area and entropy model. The landslide susceptibility maps were generated by the GIS-based algorithms, and the resultant was validated using the training (70%) and test (30%) data of landslide locations for success and prediction rates, respectively. According to the entropy model, distance to road, rainfall, and SPI are the most effective factors on landslide occurrence in the area. The area under the curve (AUC) of ROC for the ANFIS, ANFIS-ACOR, and ANFIS-DE algorithms ranges from 0.845 to 0.946 for success rate curves and 0.793 to 0.924 for prediction rate curves, respectively. Therefore, performances of the analyzed models of landslide susceptibility are good to excellent. The success rate curves suggest that the employed algorithms have high prediction performance, but the success rate curves indicate that the ANFIS-DE algorithm has the best estimation performance (0.946) with respect to the other models.

在这项研究中，结合自适应蚁群优化（ACOR）和差分进化（DE）算法，使用自适应神经模糊推理系统（ANFIS）提供了Fahliyan子盆地的滑坡敏感性图。61个滑坡中有43个（占70％）用于提供滑坡敏感性图，而18个滑坡（占30％）用于验证模型。十三种滑坡控制因素，包括海拔，平面曲率，坡度角，坡向，剖面曲率，到道路的距离，到河流的距离，到断层的距离，降雨，TWI，SPI，土地利用和岩性被用来提供滑坡敏感性图。根据滑坡相对于类别面积的频率比和熵模型，计算每个有效因子类别和有效因子的权重。滑坡敏感性图是通过基于GIS的算法生成的，并使用滑坡位置的训练数据（70％）和测试数据（30％）分别对成功率和预测率进行了验证。根据熵模型，到道路的距离，降雨量和SPI是该地区发生滑坡的最有效因素。ANFIS，ANFIS-ACOR和ANFIS-DE算法的ROC曲线下面积（AUC）分别在成功率曲线的0.845至0.946和预测率曲线的0.793至0.924的范围内。因此，滑坡敏感性分析模型的性能优良。成功率曲线表明所采用的算法具有较高的预测性能，但成功率曲线表明ANFIS-DE算法相对于其他模型具有最佳的估计性能（0.946）。