[Objective] The law of landslide deformation was effectively studied in order to produce high-precision predictions of landslide deformation. [Methods] Using the results of landslide deformation monitoring, an optimized empirical model was used to decompose deformation data. The optimized radial basis function neural network and Markov chain were then used to complete the sub-item combination prediction of landslide deformation. Finally, the seasonal Kendall test was used to judge the landslide deformation trend to verify the reliability of the prediction results. [Results] The empirical model effectively decomposed landslide deformation information, and the decomposition effect was further improved through optimization. The decomposition effect of the complementary ensemble empirical model was the best. The average relative error of the prediction results was less than 2%. This high prediction accuracy verified the effectiveness of the prediction model. The deformation trend judgment results were consistent with the prediction results, indicating that the prediction process was reliable, and that landslide deformation was increasing continuously. [Conclusion] Because landslide deformation tends to increase over time and landslide stability tends to develop in an unfavorable direction, landslide disaster prevention and control should be carried out as soon as possible.