In this study, an integrated machine learning (ML) and digital twin system was established to monitor and predict the stability of the slopes in real-time, based on the 847 monitoring samples that were taken daily from an instrumented loose deposit slope in Southwest China for about 28 months from January 2022 to May 2024. The SVM model performed best with the highest R² value among all the models, followed by the RF model and the XGBoost model, whereas the LSTM model did not perform well in the regression task with R² = −0.3675 and predictions were almost constant without capturing the temporal variations in FoS. The best overall performance (F1 = 0.955, accuracy = 96.09%) for stability classification on chronological test subset was obtained by XGBoost; SVM had a 100% recall on the marginally stable minority class, but a lower overall accuracy (85.94%). The shallow and deep pore-water pressure were the most important variables for the feature importance analysis (impurity/gain-based) and for the SHAP analysis, with 71-75% of the predictive importance. A digital twin platform has been created that provides integration of real-time sensor data, physics-based FEM complementation, quantile-based prediction intervals, and three-dimensional visualization, and can complete a full sensor-to-alert cycle in less than 30 seconds. It observed that conventional machine learning models (SVM, RF, XGBoost) can be used for prediction of the FoS for operational slope monitoring and the LSTM architecture used is not suitable for regression or safety-critical classification on this dataset.