Abstract: Accurate prediction of landslide displacement is essential for establishing effective geological hazard early warning systems. While existing research has primarily emphasized predictive accuracy,the quantification of model uncertainty has often been overlooked. To address this gap,this study proposes a novel hybrid point-interval prediction framework that integrates the Sparrow Search Algorithm(SSA),Convolutional Neural Network(CNN),Bidirectional Long Short-Term Memory(BiLSTM),Attention Mechanism,and Kernel Density Estimation(KDE). The methodology first decomposes landslide displacement into trend and periodic components using Variational Mode Decomposition(VMD). A Multilayer Perceptron(MLP)is then employed to predict the trend displacement,while the SSA-CNN-BiLSTM-AM model is used to forecast the periodic component. KDE is subsequently applied to estimate the probability density of prediction errors,generating prediction intervals at multiple confidence levels to quantify uncertainty. A case study conducted on the Bazimen landslide demonstrated the superiority of the proposed framework. Compared to a conventional LSTM model,the SSA-CNN-BiLSTM-AM model reduced the Mean Absolute Error(MAE)by 18.27% and the Root Mean Squared Error(RMSE)by 29.14%,while increasing the Coefficient of Determination(R²)by 8.91%. The Coverage Width-based Criterion(CWC)decreased by 78.008,confirming both improved point prediction accuracy and effective uncertainty quantification. These advances provide a scientifically robust foundation for enhancing landslide risk assessment and early warning capabilities,ultimately contributing to more reliable landslide hazard identification and pre-failure alert systems.