Abstract: The creep soil landslide generally goes through the initial deformation stage, the uniform deformation stage and the accelerated deformation stage from the deformation start to the overall failure. Before the instability and failure, it has obvious gradual and stage characteristics. Therefore, the surface displacement monitoring data of the creep soil landslide has the problems of excessive abnormal data and data imbalance, which affects the prediction accuracy of the model. In view of the above problems, this paper proposes a balanced optimization algorithm based on Local Outlier Factor(LOF)and Adaptive Synthetic Sampling(ADASYN)to optimize the surface displacement data of each stage of the soil landslide, and then establishes a GA-BP neural network prediction model to verify the feasibility of the LOF-ADASYN algorithm. Taking the Leigongzui landslide in Hubei Province as an example, the surface displacement data before and after the balanced optimization are input into the GA-BP model to compare the prediction results. The experimental results show that after the balanced optimization of the LOF-ADASYN algorithm, the fitting coefficient of the prediction model is increased by 0.119, the mean square error and the mean absolute error are reduced by 27.42%and 39.23%respectively. The LOF-ADASYN algorithm can solve the problems of excessive abnormal data and data imbalance in the displacement monitoring data of the creep soil landslide, and improve the prediction accuracy of the model.