Due to vibrant tectonic activity in the Taiwan area, the rock formations are highly fractured. The weathered formations or colluviums are susceptible to landslides due to the dynamic change in river morphology. The impact of climate change makes the landslide hazard more significant over the past decade. Focusing on the slate formation region in the upstreams of the Tachia River, Wu River, and Chuoshui River in Taiwan, this study explored the behavior of deep-seated and shallow landslides. In order to more accurately distinguish the deep-seated and shallow landslides, this study adopts the SHALSTAB model with the consideration of slope angle. And then, based on the classified deep-seated and shallow landslide inventories, the landslide susceptibility models, i.e., Logistic Regression and Support Vector Machine models, were created. These susceptibility models were developed for different watershed and in different scales; their performances were evaluated and compared, including the AUC values and the control factor weightings. A random sampling logic was proposed to increase the analysis efficiency, and results show a promising performance. The developed models in this study can more properly and efficiently generate susceptibility maps for landslide hazard assessment and mitigation.