<![CDATA[Landslides are one of the most frequent and destructive natural disasters in Indonesia, yet national susceptibility assessments still rely heavily on heuristic frameworks based on expert judgments. This study evaluated the performance and interpretability of two Machine Learning (ML) models, Logistic Regression (LR) and Decision Tree (DT), compared with the heuristic baseline for landslide susceptibility mapping in the Ciujung sub-watershed, West Java-Banten. The models were developed using the NASA Global Landslide Catalog and 13 conditioning factors at 30 m spatial resolution. The LR model passed the multicollinearity test and identified annual rainfall, soil texture, and land use as dominant factors, while DT identified the same factors as the main controller, but emphasized the role of local threshold and interactions between factors. Slope does not appear as a dominant factor due to the mudslide character that not only occurs on steep slopes, but also spreads to flatter areas as deposition zones, thereby reducing the discriminating power of slope in pixel-based modeling. The models of LR and DT performed quite well (AUC>0.9) and generated spatially limited susceptibility zones. In contrast to the heuristic baseline, which sometimes overestimates the hazard areas. The findings indicated that LR and DT frameworks are able to improve the methodological transparency, spatial accuracy, and policy relevance, offering a viable approach for evidence-based catastrophe governance.]]>
