Artificial Neural Network with Hyperparameter Tuning in Forecasting Scour Depths at Seawalls

Scour depth can have detrimental consequences for the structural safety and stability of coastal defence structures such as vertical walls by threatening structural integrity. The increasing frequency of storm surges caused by the changing climate calls for broader implementation of coastal defence structures and protection of existing structures from scouring. Hence, reliable predictive models for estimating scour depth are pertinent to mitigate coastal climatic hazards and infrastructure failure risks. Traditional methods for predicting scour depth have relied on empirical-based equations from experimental studies and numerical modelling methods. However, experimental methods are limited to the controlled environment, and numerical methods are complex and resource intensive. Artificial intelligence (AI) methods such as machine learning (ML) approaches have the potential to be a fast but reliable prediction tool for estimating scour depth. This study investigates the performance of the deep Artificial Neural Networks (ANN) model for predicting scour depth at vertical seawalls. The physical modelling dataset comprises experimental data of scour depths performed at the University of Warwick, UK. The effects of two hyperparameter tuning methods, namely RandomSearch (RS) and GridSearch (GS), on the performance of the deep ANN model, are evaluated. The results are compared in terms of Mean Absolute Error (MAE) and Pearson Correlation value, R. The ANN model with RS-based hypertuning yielded more accurate predictive results compared to the GS-based model with MAE and R values of 0.17 and 0.76, respectively.