Step-like displacement prediction of reservoir landslides based on a metaheuristic-optimized KELM: a comparative study

This paper applies eight state-of-the-art metaheuristic optimization algorithms to optimize the kernel extreme learning machine (KELM) and compares the mean accuracy and stability of these algorithms in step-like displacement prediction of reservoir landslides. The Baishuihe landslide, Shuping landslide, and Baijiabao landslide were selected as study cases. A new index that can balance the goodness and error of predictions, the combined prediction precision criterion (CPC), was proposed for the evaluation of these algorithms. The results show that the mean prediction accuracy between different algorithms is small, up to only approximately 15%, but the difference between the prediction stabilities varies greatly, up to approximately 70%. All metaheuristic-optimized KELM models have high mean prediction accuracy, among which the horse herd optimization algorithm is better than the other algorithms in terms of both mean prediction accuracy and stability. Our results highlight the importance of ensemble predictions in attenuating the contingency of single metaheuristic-optimized machine learning (ML) models and improving their stability and demonstrate the impartiality of the CPC index in the comparative evaluation of models. The methodological framework of this article can provide a reference for future research on ML-based landslide displacement prediction.