Abstract: In this study, we propose a dynamic prediction model based on singular spectrum analysis (SSA) and long short-term memory neural network (LSTM) to improve the prediction accuracy of landslide displacement. First, we use SSA to decompose the landslide displacement time series into a trend term and a periodic term displacement subsequence. We eliminate noise to reduce the influence of random fluctuation on experimental results. Then, we use the Gaussian fitting method to fit and predict the trend term displacement subsequence. We apply the LSTM neural network model to predict the periodic term displacement subsequence. Lastly, we obtain the prediction value of cumulative landslide displacement by superimposing the predicted values of each displacement sequence. We update the LSTM network by adding prediction values to the training set to achieve the dynamic displacement prediction. Results show that the LSTM model is better than the traditional back propagation (BP) neural network, support vector machine (SVM), and differential autoregressive integrated moving average (ARIMA). Furthermore, we verify the model in Xintan Landslide in China. The final predictive value demonstrates that the combined model has a high accuracy.