Slope stability prediction based on the combination of deep kernel learning and gaussian process

Given the complex nonlinear relationships among slope features and between those features and the factor of safety (FoS), existing machine learning–based slope stability prediction methods are limited in modeling intricate structures and fail to quantify prediction uncertainty. We propose a slope stability prediction approach that integrates deep kernel learning with Gaussian Processes. Specifically, a multilayer neural network is employed to extract highly abstract feature representations, capturing deep patterns inaccessible to shallow architectures. These latent features construct a flexible, data-driven kernel that effectively models complex nonlinear relationships underlying the slope data. Finally, a Gaussian Process utilizes this deep kernel to predict the FoS accompanied by intrinsic uncertainty quantification. Experiments on a public Kaggle dataset show that the proposed method outperforms classical machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), Gaussian Process Regression (GPR)—and deep learning methods—Gated Recurrent Unit (GRU) and Deep Neural Network (DNN)—achieving the best results in terms of root mean square error, mean absolute error, and R², thereby providing new technical support for intelligent early warning of slope hazards.