Research on landslide early warning model driven by ISSA optimization integrated machine learning algorithm

The accurate and efficient calculation of the slope stability coefficient, leading to a reliable assessment of slope stability, remains a primary goal in geotechnical engineering. However, the predictive performance of traditional models is often limited by complex factors, including variable geotechnical conditions and external environmental influences. To address this challenge, an Improved Sparrow Search Algorithm-Optimized Random Forest and Categorical Boosting Ensemble Model fusion prediction model is constructed in this study. A visualization method that combines violin and box plots is adopted to analyze data distribution, and the Pearson correlation coefficient is used to select relevant traits for dataset construction. Hyperparameter optimization of integrated models is investigated, and an enhanced Sparrow Search Algorithm (SSA) is introduced as a solution by incorporating a chaotic system to improve convergence efficiency and avoid local optima. The improved SSA is applied to optimize two algorithms Random Forest and Category Boosting Regression as base learners, with a multi-layer perceptron serving as the meta-learner, and fusion is performed through five-fold cross-validation. Verification results using actual engineering data show that the model s prediction results are in good agreement with theoretical calculation values and are significantly superior to both single models and other fusion models in terms of safety factor prediction accuracy. The model achieved excellent performance metrics, with MAE, MSE, RMSE, and R$^2$ values of 0.0586, 0.0090, 0.0949, and 0.9008, respectively. These results demonstrate the potential of the proposed framework as a robust and accurate tool for slope stability assessment.