Abstract

Ranikhola River Sub-Basin features a complex topography and frequent landslides. Road construction and urban expansion exacerbate the situation. Multi-criteria decision-making (MCDM) and machine learning methods are used to assess spatial landslides in the Ranikhola River Sub-Basin, East Sikkim. Thirteen landslide conditioning factors were represented as GIS layers in the study area to construct landslide susceptibility maps. Parameters were assigned weights using the Analytic Hierarchy Process (AHP) and Entropy methods to balance subjective and objective influences. Random forests (RFs) and artificial neural networks (ANNs) were compared with two MCDM methods, TOPSIS and VIKOR, to map landslide susceptibility. The dataset was divided into training and testing, with a 70:30 ratio. As shown by accuracy assessments, ML approaches outperform MCDM. Slope (23.21%) and rainfall (16.87%) were the most influential factors. Model validation shows that ML approaches outperform MCDM methods, with RF achieving the highest performance (sensitivity = 0.96, specificity = 0.97, F1 score = 0.96). TOPSIS and VIKOR had high specificity but low sensitivity, making them suitable for further regional studies. Although RF and ANN generally outperform MCDM methods, their comparative advantage is particularly evident in this geomorphically complex Himalayan basin, where nonlinear factor interactions are pronounced. Spatial prediction enables mitigation methods to target the most vulnerable areas of this complex landscape.