Toward the reliable prediction of reservoir landslide displacement using earthworm optimization algorithm-optimized support vector regression (EOA-SVR)

Reliable prediction of reservoir displacement is essential for practical applications. Machine learning offers an attractive and accessible set of tools for the displacement prediction of reservoir landslides. In the present study, earthworm optimization algorithm-optimized support vector regression (EOA-SVR) was proposed for the reliable prediction of reservoir landslide displacement. The proposed approach was evaluated and compared with metaheuristics, including artificial bee colony (ABC), biogeography-based optimization (BBO), genetic algorithm (GA), gray wolf optimization (GWO), particle swarm optimization (PSO), and water cycle algorithm (WCA), by the Friedman and post hoc Nemenyi tests. The results from the Baishuihe landslide showed that the EOA-optimized SVR provided satisfactory performance with a Kling–Gupta efficiency (KGE) greater than 0.98 and nearly optimal values of the coefficient of determination. Significant performance differences were revealed between the compared metaheuristics. The EOA is superior with respect to both performance and stability. The hyperparameter sensitivity analysis demonstrated that the EOA can stably provide reliable predictions by maintaining the optimal solution. The experimental results from the Baishuihe landslide indicate that the EOA-optimized SVR is promising for accurate and reliable prediction of reservoir landslide displacements, thus aiding in medium- and long-term landslide early warning.