Abstract: Rainfall is a key trigger for landslide disasters in open-pit mines, and its gradual deformation characteristics exhibit a significant coupling relationship with seasonal rainfall processes. To address the deficiencies of traditional methods in analyzing the rainfall-displacement correlation mechanism and predicting dynamic trends, a hybrid prediction model of convolutional neural network (CNN) and Kolmogorov Arnold network (KAN) integrating multi-source information is established. Taking the slope deformation of the southern dump in south Hesigewula Open-pit Coal Mine in Inner Mongolia as a background, the response relationship between slope displacement and rainfall factors is quantitatively revealed based on Pearson correlation analysis. The optimal rainfall influence period affecting the deformation of the southern slope is determined. The established prediction model is trained and validated using existing rainfall and displacement monitoring data. The long-term displacement evolution trend of the slope under different rainfall scenarios (reducing rainfall intensity by 10% and increasing rainfall intensity by 10%) is simulated at selected GNSS monitoring points. The results show that: ① compared with traditional models, the proposed CNN-KAN hybrid prediction model demonstrates superior accuracy in predicting rainfall-induced slope deformation scenarios on the same dataset; ② seasonal rainfall exhibits a lag effect on slope deformation, with the displacement-rainfall correlation coefficient at the south dumping slope decreasing below 0.2 when cumulative precipitation time reach 11 days; ③ according to the model’s prediction, when the rainfall intensity increases by 10%, the slope enters an accelerated creep stage, and the cumulative displacement value can reach 558.12 mm, and the southern slope shows a significant trend of secondary landslide development.