Prediction of Extreme Wave Heights via a Fuzzy-Based Cascade Ensemble Model

This paper addresses the critical challenge of accurately forecasting extreme wave heights, a crucial aspect for offshore operations often underexplored in existing literature. Employing a fuzzy-based cascade ensemble of regression models, our approach involves successive partitioning of training data into fuzzy-soft clusters, enabling specific regression models to analyze distinct segments of the target domain. Integration of individual model predictions into a fuzzy-based ensemble, with pertinence values assigned based on previous layer predictions, enhances accuracy by prioritizing certain events. The simplicity of our approach, eliminating the need for data balancing techniques, and its efficacy in predicting extreme wave heights with remarkable results distinguish it from existing methods. Since the optimal data partitioning is specific to the problem, an optimization strategy using two evolutionary algorithms as DE and CRO is employed to determine specific parameters of the methodology, including the number of membership functions, shapes of membership functions, and learning rate. This optimization strategy further enhances its performance, making it a promising solution for wave forecasting challenges.