Qiongjie County, located along the southern margin of the Qinghai-Tibet Plateau, is highly susceptible to geological hazards due to its complex tectonic setting, frequent seismic activity, abundant precipitation, and increasing anthropogenic disturbances. To quantitatively assess landslide susceptibility across the region, we developed a multi-factor evaluation framework integrating topographic, geological, climatic, hydrological, vegetation, and anthropogenic variables derived from field investigations and multi-source remote sensing data. By coupling the Information Value (IV) model with machine-learning algorithms, namely Random Forest (RF) and Convolutional Neural Networks (CNN), two hybrid susceptibility models (IV-RF and IV-CNN) were constructed. The hybrid models exhibited substantially improved predictive performance relative to the individual algorithms, achieving AUC values of 0.965 for IV-RF and 0.976 for IV-CNN, compared with 0.859 for RF and 0.879 for CNN. Susceptibility mapping based on the IV-CNN model indicates that 12.76%, 9.48%, 8.47%, 12.27%, and 57.03% of the study area fall within extremely high, high, moderate, low, and extremely low susceptibility classes, respectively. Areas of elevated landslide risk are primarily concentrated around Qiongjie Town, Jiama Township, Layu Township, and Xiashui Township, as well as along the Zeyu transportation corridor and tributaries of the Yarlung Tsangpo River. Factor importance analysis reveals that elevation (28.3%), annual precipitation (22.1%), distance to roads (18.7%), distance to rivers (15.4%), and vegetation cover (12.5%) are the dominant controls on landslide occurrence. Overall, this study advances the understanding of landslide susceptibility in tectonically active plateau regions and demonstrates the effectiveness of hybrid frameworks that integrate physically interpretable statistical models with data-driven machine-learning approaches for regional-scale geohazard assessment.