The study aims to develop a hybrid model approach for the assessment of the landslide (LS) risk qualitatively. It involves multiple consecutive stages of statistical prediction, machine learning, and mapping in the GIS environment. At the first stage, a landslide susceptibility map has been developed using the analytic hierarchy process (AHP) algorithm, coupled with the binary logistic regression (BLR) technique. The AHP model incorporates 11 geo-hydrological and environmental variables as predictors sourced from remote-sensing datasets to generate the LS susceptibility as output. Twenty-three field-based validation locations validate the test result. Pearson's correlation coefficient (r) between the observed (\({{\mathrm{\L}}}_{{{\text{COMPUTED}}}}\)) and predicted (\({{\mathrm{\L}}}_{{{\text{PREDICTED}}}}\)) values of LS susceptibility is 0.928 at 0.01 level of significance. At the next stage, the LS risk is evaluated considering the ‘risk trio,’ i.e., the combination of the hazard, exposure, and vulnerability. This stage involves the transformation of a range of qualitative datasets to the virtual workspace of machine learning. The landslide risk output has been predicted with an initial fuzzy model, incorporating a set of 32 rules for membership functions (MF). This initial model uses randomly selected 20% datasets to tailor the fuzzy rules through the adaptive neuro-fuzzy interface (ANFIS). The training to ANFIS results in framing 120 fuzzy rules for the best possible prediction of the outcome. The final LS risk map from the ANFIS output shows that more than 70% area is under high-to-very high LS risk. The model is tested in a 5′ × 5′ grid around the famous hill station Mussoorie in the state of Uttarakhand, India. The model exhibits a satisfactory level of accuracy for the present-study area, which has made us confident to recommend it. The multistage model is worthy of being applied for landslide risk mapping for the similar kinds of study areas, and also for other areas of landslide with necessary customization as deemed necessary.

该研究旨在开发一种混合模型方法，用于定性评估滑坡（LS）风险。它涉及GIS环境中的统计预测，机器学习和制图的多个连续阶段。在第一阶段，已经使用层次分析法（AHP）算法以及二进制逻辑回归（BLR）技术开发了滑坡敏感性图。AHP模型结合了11个地理水文和环境变量作为从遥感数据集获得的预测因子，以生成LS敏感性作为输出。23个基于现场的验证位置可验证测试结果。观察到的（\（{{\ mathrm {\ L}}} _ {{{\ text {COMPUTED}}}}} \）和预测值之间的Pearson相关系数（r）\（{{\ mathrm {\ L}}} _ {{{text {PREDICTED}}}} \\））在0.01的显着性水平下LS敏感性值为0.928。在下一阶段，将根据“风险三重奏”（即危害，暴露和脆弱性的组合）评估LS风险。此阶段涉及将一系列定性数据集转换为机器学习的虚拟工作区。滑坡风险输出已通过初始模糊模型进行了预测，并纳入了隶属函数（MF）的一组32条规则。该初始模型使用随机选择的20％数据集来通过自适应神经模糊接口（ANFIS）定制模糊规则。对ANFIS的训练导致为最佳结果的最佳预测构图120条模糊规则。来自ANFIS输出的最终LS风险图显示，超过70％的区域处于高到非常高的LS风险之下。该模型在印度北阿坎德邦州著名的小山站Mussoorie周围的5'×5'网格中进行了测试。该模型在当前研究领域显示出令人满意的精度水平，这使我们有信心推荐该模型。该多阶段模型值得用于类似研究区域的滑坡风险地图绘制，以及值得认为必要的必要定制的其他滑坡区域地图。