.png){kind=icon}
.png){kind=icon}
Return
Hyperparameter-tuned Light Gradient Boosting Machine model for predicting breaking wave height
.png){kind=icon}
PRE
AI
DOI：10.1080/21664250.2025.2605832.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study proposes a new model using Light Gradient Boosting Machine (LightGBM) to predict breaking wave height based on input wave parameters. To determine optimal hyperparameter, Optuna is employed and conducts 100 independent runs with 10-fold cross-validation. Additionally, SHapley Additive exPlanations (SHAP) analysis is applied to investigate behavior of model. Results show that LightGBM model optimized with Optuna shows excellent performance for estimating breaker height. Root mean square error of model is 1.861 cm (for training dataset) and 3.518 cm (for testing dataset). Coefficients of determination are also high with 0.998 and 0.992 for training and testing datasets, respectively. This accuracy is remarkably higher than previous existing breaking wave height models. Besides, SHAP analysis highlights that deep-water wave height and water depth have the greatest impact on breaker height prediction. The results demonstrate that combination of Optuna and LightGBM enhances robustness and generalization of model for predicting breaking wave height.
Keywords:
Breaking wave height
Light Gradient Boosting Machine
Optuna
SHapley Additive exPlanations
Journal
C
IF:
1.9
Papers:
31
Citations:
0