Abstract
In this study, artificial neural networks (ANN) and multiple additive regression trees (MART) are employed to predict significant wave heights (Hs) based on the measurement of meteorological and wave data. Four ANN models are proposed, namely, the multilayer perceptron neural network (MPNN), cascade correlation neural network (CCNN), radial basis function neural network (RBFNN), and general regression neural network (GRNN). The inputs to the various models include wind speed, wind direction, wind duration, fetch, sea level pressure, and air temperature. The models are trained and tested by the cross-validation method and compared for performance outcomes based on six different performance measures. Analyzing the ANN models, the overall performance of the RBFNN is found to be more accurate than those of the other models, and the CCNN model exhibits the worst predictive capabilities. The MART model is the most accurate, outperforming all other models based on the statistical indices. The MART model is efficient and precise and can, therefore, serve as a practical tool for Hs prediction.
•Using MART and NNs to predict wave height based on measuring meteorological and wave data.•For ANN models, the performance of the RBFNN was superior, while the CCNN exhibits the worst predictive capabilities.•The MART model was the most accurate, outperforming in all statistical indices' terms compared with other models.•The study helps with the design and operation of coastal and offshore structures.