A Deep Learning-Based Prediction Model for the Energy Dissipation Level of Shear Walls

Abstract

This study explores the application of deep learning for predicting the energy dissipation level of reinforced concrete shear walls. Traditional empirical formulas and conventional machine learning methods often face limitations in accuracy and generalizability when dealing with complex, high-dimensional data. To address these challenges, a deep neural network model is constructed and trained on a comprehensive dataset comprising 312 shear wall specimens with 21 feature parameters. Through comparative experiments with five machine learning models—Ridge Regression, Lasso Regression, Random Forest, Gradient Boosting, and XGBoost—the deep learning model demonstrates superior performance, achieving a higher coefficient of determination ( ) and lower prediction errors (MSE and MAE). Feature importance analysis via Pearson correlation further refines the input variables, identifying 12 key parameters that significantly influence energy dissipation capacity. The results validate the potential of deep learning as a robust and efficient tool for seismic performance assessment of shear walls.