Trends and insights in machine learning-based landslide susceptibility research and early warning practices: a bibliometric analysis

Landslides are naturally occurring phenomena that pose significant threats to human livelihoods and may fully or partially damage the environment and infrastructure. With climate change and increasing extremes, landslide events have become more frequent, highlighting the need for advanced prediction, monitoring, and risk assessment approaches. This study presents a bibliometric analysis of the landslide research from 2016 to 2023, focusing on the use of machine learning (ML) techniques. A total of 3043 articles from 73 Web of Science (WoS) journals by 8051 authors across different research global institutions were analysed and findings were validated using Scopus database for the same period. The results show a consistent increase in the number of publications using artificial intelligence (AI) techniques for landslide studies, where China emerged as the leading contributor, followed by Italy, the USA, and India. The average citation per article suggests the field is still emerging and application-oriented, and its influence is yet to be fully captured by traditional citation indicators. Debris flows, slope stability, and susceptibility were frequently studied themes, which coalesced into six research clusters comprising geotechnical processes, hazard assessment, methodological advances, environmental controls, monitoring approaches, and risk management. Beyond 2020, the traditional geotechnical methods maintained a stable growth, while advanced computational approaches, including ML, deep learning (DL), and interferometric synthetic aperture radar (InSAR), showed increased adoption, indicating an evolving methodological trend. The meta-analysis revealed slope as the most frequently studied landslide conditioning factor (∼1500 occurrences), while notable gaps remain in understanding anthropogenic factors and human-environment interactions. Logistic Regression is the most frequently used method, followed by Support Vector Machine, and Random Forest indicating a preference for models that balance predictive performance and interpretability. Alignment with the Sustainable Development Goals mainly focused on Infrastructure (60.3%) and Climate Action (36.2%), with limited representation in ecosystem impacts and public health, revealing opportunities for more integrated approaches that bridge methodological advances and real-world application.