Abstract: Accurate prediction of landslide displacement is a crucial component of landslide early warning systems. This paper proposes a landslide displacement prediction model based on Gaussian Process Regression (GPR) combined with diverse time-series feature engineering, achieving high-precision displacement prediction and uncertainty quantification. TAKING THE BAZIMEN LANDSLIDE AS AN EXAMPLE, During the feature engineering phase, displacement lag features, rolling mean of rainfall, rolling variance of reservoir water level, and displacement change rate are constructed. Additionally, temporal decomposition features including monthly and quarterly components are extracted. SUBSEQUENTLY, EMPLOY THREE-FOLD TIME SERIES CROSS-VALIDATION, ALONG WITH A GRID SEARCH SCHEME, TO OPTIMIZE HYPERPARAMETERS IN CONJUNCTION WITH THE TIME SERIES CROSS-VALIDATION STRATEGY, THEREBY MITIGATING THE RISK OF OVERFITTING IN THE SMALL-SAMPLE SCENARIO. The results demonstrate that after incorporating multi-source temporal features, the prediction coefficients of determination (R²) for monitoring points ZG110 and ZG111 at the Bazimen Landslide significantly increase to above 0.99. Metrics such as MAE, RMSE, and MAPE are substantially reduced, indicating a significant improvement in prediction accuracy. This study integrates probabilistic modeling with feature interpretability analysis. The proposed method achieves high-precision landslide displacement prediction in small-sample environments while simultaneously quantifying prediction uncertainty. It provides effective decision support for landslide risk early warning and engineering safety assessment.