A Transferable Machine Learning Approach for Identifying Rainfall-Induced Cliff-Type (Shallow) Landslides in Seismic and Non-Seismic Regions

Precise classification of landslide types is essential for effective hazard mitigation; however, many existing landslide inventories lack type-specific information, limiting their applicability in risk management. This study presents a transferable machine learning framework to identify rainfall-induced cliff-type landslides—commonly corresponding to shallow landslides in Japan—from unclassified inventories across both seismic and non-seismic environments. Using Forest-based and Boosted Classification and Regression (FBCR) tools in ArcGIS Pro, the model was developed based on 25 landslide conditioning factors using balanced datasets of cliff-type and non-landslide samples derived from Tokushima and Wakayama Prefectures, Japan.The model achieved strong predictive performance in the training regions, with accuracy and sensitivity exceeding 0.84, an F1 score of approximately 0.84–0.85, and a Matthews correlation coefficient (MCC) ranging from 0.68 to 0.71. Transferability was evaluated by applying the trained model to the Kegalle District in Sri Lanka, where it achieved an accuracy of approximately 80% against available inventory data. Variable importance analysis revealed that rainfall consistently ranks among the most influential triggering factors for cliff-type (shallow) landslides, even in earthquake-prone regions, where seismic-related variables exhibited comparatively lower influence. Key controlling factors included rainfall, slope, elevation, proximity to infrastructure, and hydrological indices.These findings highlight that rainfall remains a dominant trigger for shallow landslides across different tectonic settings. The proposed framework provides a practical approach for complementing missing landslide type information in existing inventories, thereby improving hazard zonation and supporting risk-informed planning in diverse environmental conditions.