Improved data-driven models for estimating shear capacity of squat rectangular reinforced concrete walls

Abstract

Reinforced concrete (RC) shear walls are very commonly used in buildings and nuclear power plants. Shear strength is one of the critical parameters in the design of RC walls, especially considering the influence of horizontal loads such as wind or earthquakes. The objective of this paper is to build machine learning (ML) models to predict the shear capacity of rectangular RC squat walls. A dataset of 312 experimental results in previous studies were collected and used for training ML models. Six ML models including Artificial neural network-Levenberg Marquardt (ANN-LM), Artificial neural network-Bayesian regularization (ANN-BR), Artificial neural network-Gene algorithm (ANN-GA), Adaptive neuro fuzzy inference system (ANFIS), Random Forest (RF), and Gradient boosting regression tree (GBRT), were developed to predict the shear strength of RC walls. The prediction results of the proposed ML models were compared with that from eight empirical formulas in design standards and published studies. From the comparison results, the RF and GBRT models predicted the shear capacity of RC walls much more accurately than existing formulas. Furthermore, a graphical user interface has been established based on an efficient ML model to facilitate the actual design process of rectangular RC short walls.