Unveiling the layers of decision-making in landslide susceptibility: a critical exploration of multi-criteria analysis, bivariate statistics, and knowledge-based approaches for complex relationships of factors and their weightage

Landslides represent one of the most destructive forms of natural hazards, often resulting in severe threats to life, infrastructure, and the environment. The accurate identification and evaluation of landslide conditioning factors (LCFs) are therefore critical for the development of reliable susceptibility maps. However, traditional approaches often assess each factor in isolation, failing to capture the complex and interdependent relationships that exist among them. These limitations reduce the scientific robustness and predictive precision of the resulting susceptibility models. In this study, we address these shortcomings by knowledge-driven approach to assign weights and ranks to 15 LCFs. This methodology enables a clearer understanding of the intricate interactions among variables such as elevation, distance to road, and proximity to fault lines that collectively influence landslide occurrences. The derived weights, calculated using Information Gain (IG), reflect the true influence of each factor in a multi-dimensional context, rather than through isolated statistical correlations. These refined weights were then integrated into classical modelling frameworks, namely Multi-Criteria Decision Making (MCDM) and Bivariate Statistical (BV) analysis. A comparative evaluation demonstrated a significant enhancement in model performance following the integration of knowledge-based weights. The accuracy of the susceptibility maps increased by over 25% for MCDM and 7% for BV models, validating the advantage of using knowledge-driven approach to uncover latent patterns and dependencies among factors. This approach may enhance predictive reliability and enables more precise identification of high-risk zones, supporting effective mitigation and disaster planning.