<p>This study presents a comparative assessment of landslide susceptibility in the Piryu River Basin, Democratic People’s Republic of Korea (DPRK), via three modeling approaches: the analytical hierarchy process (AHP), frequency ratio (FR), and Shannon’s entropy (SE). Eleven conditioning factors—slope, aspect, land use, curvature, lithology, precipitation, distances to faults, rivers, and roads, as well as the stream power index (SPI) and topographic wetness index (TWI)—were derived from remote sensing data, geological maps, and meteorological records. A total of 100 landslide points were identified in the study area. The AHP method was applied on the basis of expert judgment through pairwise comparisons, whereas FR and SE were used as data-driven benchmark models. Validation via the ROC-AUC, confusion matrix, and kappa coefficient revealed that the SE model achieved the highest predictive performance (AUC = 0.8589, accuracy = 85.00%, kappa = 0.6000), followed by FR (AUC = 0.8200, accuracy = 80.00%, kappa = 0.6000) and AHP (AUC = 0.8200, accuracy = 78.33%, kappa = 0.5667). The AHP model classified 14.04% as very low, 20.29% as low, 25.14% as moderate, 25.79% as high, and 14.74% as very high. The FR model classified 14.48% as very low, 30.56% as low, 31.61% as moderate, 17.59% as high, and 5.77% as very high. The SE model classified 10.96% as very low, 26.20% as low, 35.06% as moderate, 20.39% as high, and 7.38% as very high. The high- and very high-susceptibility zones are predominantly located in steep mountainous areas underlain by weathered sedimentary rocks. The findings demonstrate that data-driven approaches outperform the expert-driven AHP model, although the AHP still provides acceptable results. The integrated framework offers a replicable methodology for landslide susceptibility mapping in data-scarce regions.</p>