Author(s): M. A. Habib; S. Abolfathi; J. J. O’Sullivan; M. Salauddin
Linked Author(s):
Keywords: Rtificial intelligence; Machine learning; Overtopping; ANN; GBDT; Random forest; SVMR; Sloping structures
Abstract: The prediction of wave overtopping at coastal defenses is critical to ensure the flood resilience of people and properties in low-lying nearshore coastal areas. With the effects of anthropogenic climate change, the frequency of wave overtopping is expected to increase, along with sea level rise and more frequent damaging storm surges. Established approaches for the prediction of wave overtopping have traditionally relied on physical and numerical modelling and empirical methods. The ubiquity of computational resources has led to the emergence of Artificial Intelligence techniques, such as Machine Learning (ML) algorithms, as a promising approach for predicting wave overtopping. This study investigates the application of four ML models based on Random Forest (RF), Gradient Boosted Decision Trees (GBDT), Support Vector Machines Regression (SVMR) and Artificial Neural Networks (ANN) approach for predicting wave overtopping at sloping breakwaters. Data from the EurOtop II manual, a comprehensive dataset of physical and numerical wave overtopping tests undertaken on a variety of coastal structure geometries, including sloping breakwaters (the focus of this study), underpinned the developed models. . To optimize the data for redundancy, feature transformation and advanced feature selection methods were employed. Hyperparameter tuning was performed to extract the best features for the predictive models. The performance of the developed ML-based models was examined in terms of the coefficient of determination, r2, and the Pearson correlation coefficient, R, for the measured and predicted overtopping values. The range of r2 values across the four models varied between 0.69 to 0.87, with Pearson correlations varying between 0.87 and 0.93. The results show that the GBDT model outperformed the other ML models tested in this study.
DOI: https://doi.org/10.3850/978-90-833476-1-5_iahr40wc-p0115-cd
Year: 2023