A comparative analysis of AHP, FR, AHP‐FR and LR models for landslide susceptibility mapping in Sikkim Himalaya, India

Abstract Landslides are among the most recurrent and destructive geohazards in mountainous regions, posing serious threats to human life, infrastructure and socio‐economic stability. The Himalayan state of Sikkim, India, is particularly vulnerable due to its fragile geology, steep terrain, intense monsoonal rainfall, active tectonics and increasing anthropogenic pressures. While numerous landslide susceptibility mapping (LSM) studies exist for the Himalaya, many rely on routine model comparisons with limited methodological advancement and insufficient validation in data‐scarce environments. To address this gap, the present study systematically evaluates deterministic, statistical and hybrid modelling frameworks for LSM. Traditional and hybrid models, such as analytical hierarchy process (AHP), frequency ratio (FR), logistic regression (LR) and AHP–FR, were applied using a comprehensive inventory of 211 landslides, with 70% (148 events) used for training and 30% (63 events) for validation. Fourteen geo‐environmental conditioning factors were optimized through multicollinearity diagnostics to enhance model robustness. The resulting LSM maps were classified into five susceptibility classes, where the ‘very high’ susceptibility zones occupied 9.78% (AHP), 12.58% (FR), 16.66% (AHP–FR) and 22.92% (LR) of the study area. Model performance evaluation using seed cell area index (SCAI), success rate curve (SRC) and prediction rate curve (PRC) indicates that the hybrid AHP–FR model achieved the highest predictive accuracy (success AUC = 0.85; prediction AUC = 0.81), along with high sensitivity (0.79) and strong spatial reliability (SCAI = 0.38), followed closely by LR (prediction AUC = 0.79). The results demonstrate that structured hybridization effectively reduces subjectivity while improving spatial predictability, outperforming stand‐alone heuristic and statistical approaches. The proposed LSM framework provides practical guidance for slope stabilization, infrastructure development and disaster‐resilient land‐use planning, highlighting the suitability of moderately complex hybrid models for hazard assessment in data‐limited Himalayan regions.