Slope stability prediction is the foundation of slope disaster prevention and control. Traditional methods of slope stability prediction are mainly based on the limit equilibrium method of statics balance theory and numerical analysis methods based on elastoplastic mechanics. Such methods are characterized by large computational complexity, convergence problems, and limitations in expressing the nonlinear relationships among different components of slope systems, which remain mainly in the scientific research level. In recent years, with the development of computer networks and artificial intelligence, machine learning intelligent decision-making systems, combined with swarm intelligence optimization algorithms, have provided beneficial methods for solving problems that are complex, nonlinear and multidimensional, and have opened up new directions for slope stability prediction. Jin et al. ^[1] used sparrow search algorithm (SSA) to optimize support vector machine (SVM), and constructed an SSA-SVM model for intelligent prediction of slope instability. Li et al. ^[2] constructed an SSA-ELM open-pit slope displacement prediction model based on an improved extreme learning machine (ELM) and conducted engineering applications, indicating that the SSA algorithm has good improvement effects.

In summary, some scholars have achieved some results in the prediction of slope stability by using sparrow search algorithm. However, there are still some urgent problems that need to be solved: (1) Although some scholars have used SSA to optimize the hyper-parameters of classifiers and achieved improvement in slope stability prediction, swarm intelligence optimization algorithms still have common defects that need improvement. (2) As a swarm intelligence optimization algorithm, SSA has common defects with other swarm intelligence optimization algorithms when solving complex optimization problems. When the search approaches global optimality, the diversity of population significantly decreases, and convergence speed is not fast enough, leading to potential risks of falling into local optima. These problems will to some extent affect the final performance and prediction results of the model.

In view of above analysis, this paper proposes a sparrow search algorithm (MISSA) based on multi-strategy improvement to optimize the support vector machine (SVM) for slope stability prediction. Through one-dimensional SPM mapping, adaptive weight-cosine algorithm, Levy flight mechanism, and dynamic adjustment strategy for step size factor, the population is initialized to increase population diversity while constantly updating the position of discoverers, followers, and watchers. Multiple strategies are used to optimize and improve the slow convergence speed, low accuracy, and tendency to fall into a local optimum of the traditional SSA. Ninety sets of samples are used to compare and analyze the performance of the predictive models established in this paper with those of the SSA-SVM and SVM-BP models. The results show that the accuracy, precision, Recall, F1 score, and AUC of the MISSA-SVM model are 0.963, 1, 0.917, 0.957 and 0.958, respectively, which are better than the two comparative models. The multi-strategy optimization improvement of SSA in the optimization of the two hyperparameters c and g in SVM model has better effects. Finally, to further verify the superiority of the model established in this paper, the MISSA-SVM model, SSA-SVM model, and SVM-BP model are used to predict nine engineering examples. The results show that the predictive results of the MISSA-SVM model established in this paper are completely consistent with the actual situation, while the SSA-optimized SVM model and BP model both have three misclassified samples. These results further demonstrate the reliability and effectiveness of the MISSA-SVM model.