Based on six periods of landslide inventory data in Yunnan Province from 1986 to 2016, a weighted information value-logistic regression (WIV-LR) model integrated with geodetector was developed, and the spatio-temporal evolution of the resulting landslide susceptibility maps was analyzed. The results indicate that: 1) WIV-LR model demonstrated good predictive performance for all six periods, with AUC values ranging from 0.781 to 0.873. 2) The spatio-temporal evolution of landslide susceptibility exhibited clear and systematic patterns, characterized by a continuous expansion of high-susceptibility areas (with their proportion increasing from 35.52% to 50.27%), a concentration of high-susceptibility zones in western, central, and southern Yunnan, and a regular migration of the susceptibility centroid, showing a southward shift, followed by an eastward movement and a subsequent return toward the northwest. 3) Geodetector analysis revealed the driving mechanisms underlying these evolutionary patterns, indicating that the most influential factors exhibited stage-dependent characteristics, with seismic intensity playing a dominant role during the T1 period, rainfall exerting a stronger influence during the T4 period, and the impact of human activities becoming increasingly significant in later periods. 4) The spatio-temporal evolution of landslide susceptibility exhibited a high degree of consistency with the dynamic changes in key influencing factors. These findings provide a scientific reference for the dynamic identification and prevention of landslide risks in regions characterized by similarly complex geological environments.