2024 Volume 80 Issue 17 Article ID: 24-17028
Wave height prediction along coastlines is essential for marine construction and leisure safety. Operations rely on wave heights being under 1 meter, emphasizing the need for more accurate predictions for safety and cost-effectiveness. Recently, the use of machine learning, especially methods like XGBoost and TabNet that include feature importance evaluation, has become prominent. This study aims to develop a week-ahead wave height prediction system using these models, focusing on the impact of wind speed data variability on accuracy. It uses long-term wave and extensive wind speed observations, assessing data with feature evaluations and spatial distribution. Findings show that choosing wind speed points carefully improves prediction accuracy, especially for short-term forecasts. The study also explores the connection between accuracy and prediction period through binary classification in statistical analysis.
