Development of XGboost model for wave overtopping using enhanced clash database / Mohamed Tarek Mohamed Fouad

Mohamed Tarek , Mohamed Fouad (2024) Development of XGboost model for wave overtopping using enhanced clash database / Mohamed Tarek Mohamed Fouad. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (117Kb)
	PDF (Thesis M.A) Download (2001Kb)

Abstract

The accurate prediction of wave overtopping is crucial for designing resilient coastal structures. This thesis presents a comprehensive study on estimation of wave overtopping using (XGB) algorithm, with a focus on both model development and experimental validation. In the first part of the thesis, the focus was on the development of the XGB model for wave overtopping prediction. The methodology started with exploring the database parameters, followed by rigorous data preprocessing to ensure data quality. The model tuning process was elaborated, incorporating the utilization of hyperparameters to enhance predictive performance. After the preprocessing phase, the number of parameters chosen for the model development was 36 parameters, while the number of data points taken from the dataset was 5670 tests. The preprocessed database was split into 70% for training and 30% for testing in the XGB model. The model attained high predictive accuracy with RMSE of 0.28 m3/s/m, a percentage error of 4.9%, and a high correlation coefficient (R) of 0.95. Percentage error was used as the primary error metric, underpinning its effectiveness in quantifying differences in prediction. The thesis examined model performance in different conditions by categorizing wave overtopping rate (q) data into low, medium, and high ranges. The low range consisted of 893 points while the medium and high range contained 772 and 36 points respectively. RMSE values for low, medium, and high q ranges were 0.34 m3/s/m, 0.23 m3/s/m, and 0.17 m3/s/m, respectively. The percentage error statistics for these ranges were 4.9%, 4.9%, and 7.4%, respectively. Model validation is executed via the bootstrap resampling technique to reveal the model inherent robustness. Following the implementation of the resampling technique, the model showed a poorer result, with an RMSE of 0.31 m3/s/m, an R value of 0.94, and a percentage error of 5.4%. To validate the performance of the model, the results were compared to an existing XGB model developed by Den Bieman (DB) that used the same database. Achieving similar results confirmed the good performance of the model and the XGB technique reliability. The second part of the thesis delved into the experimental aspect, contributing novel data to the existing database. A thorough designed experiment was conducted within the National Hydraulic Research Institute of Malaysia (NAHRIM), featuring comprehensive information about the wave flume, wave generator system, and data acquisition setup. The experimental design, encompassing wave conditions and data collection procedures, was outlined. Adding 49 new tests to the existing database had a small impact on predictive performance, with a percentage error of 10.09% for the original dataset and 10.43% for the updated dataset. The combination of model development and physical experiment contributed to a better understanding of wave overtopping phenomena. The results underscored the potential of the XGB algorithm in accurate wave overtopping prediction, while also emphasized the challenges and considerations when integrating experimental data into existing predictive frameworks.

Actions (For repository staff only : Login required)