Data-driven slope stability prediction based on Gaussian process

Slope stability prediction is essential for landslide early warning and engineering risk management. However, current machine learning methods often face practical limitations, such as a lack of decision transparency and the inability to provide explicit uncertainty quantification. To address these issues, this study proposes a Gaussian process classification (GPC)-based method for slope stability prediction with an additive composite kernel. The kernel combines an automatic relevance determination (ARD) radial basis function (RBF) kernel, an ARD Matern-2.5 kernel, and a linear kernel to capture smooth nonlinear, local heterogeneous, and linear relationships among slope features. To evaluate the proposed method, 887 slope cases were compiled from published studies, with six representative features used as inputs: slope height, slope angle, unit weight, cohesion, internal friction angle, and pore water pressure ratio. The model was evaluated using 10 random 8:2 train-test splits, with five-fold cross-validation conducted on each training set for hyperparameter optimization. Compared with the support vector machine (SVM), random forest (RF), decision tree (DT), neural network (NN), and gradient boosting machine (GBM) models, the proposed GPC achieved the best average test performance, with an AUC of 0.921 ± 0.019, accuracy of 0.858 ± 0.023, precision of 0.859 ± 0.033, recall of 0.839 ± 0.045, and F1-score of 0.848 ± 0.030. These results demonstrate that the proposed method is an effective and uncertainty-aware tool for slope stability prediction, with potential for landslide monitoring and risk-informed decision-making.