Review Paper:
A review paper
on Landslide Susceptibility Mapping using Geospatial Technology and Machine Learning
Techniques
Kollipara V. Ramana, Chitikela N. Vara Laxmi, Kodimela Anil, Jampana N.S. Suryanarayana
Raju and Parupalli Sridhar
Disaster Advances; Vol. 18(11); 91-98;
doi: https://doi.org/10.25303/1811da091098; (2025)
Abstract
Landslides are among the most frequent and devastating natural hazards, often resulting
in significant loss of life, property damage and disruption to infrastructure and
agriculture. As a serious geo-environmental issue, landslides present complex challenges
for both prediction and control. Landslide Susceptibility Mapping (LSM) has emerged
as a valuable tool for identifying high-risk areas and supporting disaster mitigation
strategies. In recent years, numerous researchers have applied geospatial technologies
in combination with statistical methods and machine learning techniques to enhance
the accuracy of LSM. Review papers play a crucial role in helping researchers and
academicians to identify knowledge gaps and to evaluate existing methodologies by
synthesizing findings from previous studies. This review is based on a comprehensive
collection of research studies focused on LSM using geospatial and machine learning
approaches, aiming to provide insights into current practices and future research
directions. The analysis reveals that machine learning models, particularly Random
Forest (RF), Support Vector Machine (SVM) and Gradient Boosting Decision Trees (GBDT),
consistently outperform traditional statistical methods like Logistic Regression
(LR) and Frequency Ratio (FR) in predictive accuracy.
Studies have reported AUC values exceeding 0.95 for RF models, indicating excellent
predictive capabilities in various geographical contexts. Furthermore, the integration
of Bayesian optimization techniques has enhanced model performance, with improvements
in prediction accuracy up to 7% for GBDT models. Hybrid models, combining algorithms
such as SVM with metaheuristic optimization methods, have also demonstrated superior
performance, effectively capturing complex, nonlinear relationships inherent in
geospatial data. In conclusion, the adoption of advanced machine learning and hybrid
models has significantly improved the accuracy and reliability of LSM. These methodologies
offer robust tools for disaster risk management, enabling more effective identification
of high-risk areas and informing mitigation strategies. Future research should focus
on enhancing model interpretability and integrating real-time data to further refine
susceptibility assessments and support proactive landslide risk reduction efforts.
