Improving Bridge Safety: A Spider Monkey Optimization-based ANN Model for Scour Depth Prediction

Abstract Hydraulic engineering research has long focused on understanding and predicting scour depth around bridge piers, a critical factor in maintaining structural integrity of bridges. This study delves into applying soft computing methods, specifically machine learning algorithms, to model and simulate local scour depth around simple piers. Leveraging a robust dataset compiled from various sources and utilizing five distinct models, including Artificial Neural Networks (ANN), Gradient Tree Boosting (GTB), and CatBoost Regression (CBR), the research aims to accurately predict pier scour depth and assess the impact of different variables on the estimation process. Additionally, to enhance estimation accuracy, the neural network weights were optimized using the Spider Monkey Optimization (SMO) and Particle Swarm Optimization (PSO) methods. Using mutual information (MI) as a feature selection method, the study reveals the critical role of specific features in enhancing the precision of scour depth predictions. Through a comprehensive analysis of model performance metrics, the study highlights the efficacy of the SMO-based ANN model for accurately predicting scour depth. Furthermore, through a detailed evaluation using the Taylor diagrams, the study provides an insightful comparison of the predictive capabilities of the hybrid machine learning models, shedding light on their respective errors and accuracy in estimating scour depth around bridge piers.