Conventional and Data-Driven Models for Landslide Susceptibility Prediction: A Comprehensive Review

Abstract

Landslide susceptibility mapping (LSM) is essential for hazard assessment and mitigation, where anthropogenic, climatic and geomorphic factors can intensify instability. This review synthesizes conventional approaches—including statistical methods (e.g., logistic regression, weights of evidence), heuristic models (e.g., analytical hierarchy process, fuzzy logic), and physically based methods—with advancements in data-driven techniques such as machine learning (ML), deep learning (DL), and explainable artificial intelligence (XAI). Case studies from diverse regions, including Turkey, China, Vietnam, and India, highlight modeling applications and challenges. Conventional models offer interpretability and practicality in data-scarce settings, while ML and hybrid frameworks (e.g., random forest, ANFIS, GAMI-net) provide superior predictive capabilities, albeit with reduced transparency and higher computational demands. The review emphasizes model validation, uncertainty quantification, standardized landslide databases, and cross-regional adaptability. Emerging trends focus on integrating remote sensing with explainable AI and hybrid models to enhance LSM accuracy, transparency, and relevance for disaster management and urban planning.