Slope stability prediction under seismic loading based on the EO-LightGBM algorithm

Abstract

Predicting the stability of slopes under seismic conditions is critical for geological hazard prevention and infrastructure safety. This study proposes an optimized prediction model based on EO-LightGBM to enhance the accuracy of slope stability assessment. A dataset containing 96 numerical simulation cases was constructed using FLAC3D, incorporating key influencing factors such as slope angle, inclination angle, slope height, rock mechanical parameters, and hard-to-soft rock thickness ratio. The dataset was split into a training set (76 samples) and a test set (20 samples). LightGBM, a gradient boosting decision tree (GBDT) model, was initially trained on the dataset, while Equilibrium Optimizer (EO) was utilized for hyperparameter optimization, focusing on learning rate, number of decision trees, maximum depth, and number of leaf nodes. The five-fold cross-validation approach was adopted to evaluate model generalization ability. The experimental results demonstrate that EO-LightGBM achieves a prediction accuracy of 94.0%, precision of 96.0%, recall of 92.0%, and an F1-score of 96.0%, outperforming traditional machine learning models such as SVM, KNN, and Decision Tree. Comparative analysis further confirms that EO-LightGBM effectively reduces error rates and enhances the adaptability of slope stability prediction models under complex seismic conditions. This study provides a reliable computational tool for seismic slope stability evaluation, contributing to improved risk assessment in geotechnical engineering.