Joint shear strength prediction of beam-column connections using machine learning via experimental results

Machine learning (ML) had already advanced rapidly in recent years, promising to completely change and enhance the function of data science in a spectrum of areas. This paper proposed a framework for predicting the shear strength of reinforced concrete beam-column connections subjected to cyclic loading. Six classical prediction models (Ordinary least Squares (OLS), Support Vector Machines (SVM), K-Nearest neighbour regression (KNN), Multivariate Adaptive Regression Splines (MARS), Artificial Neural Networks (ANN), and kernel regression with mixed data types (Kernel regression)) were proposed, and a total of 98 dataset results were collected and used to train and evaluate the suggested framework models. The most important factors influencing the joint shear strength were chosen based on the previous experimental results to create a parametric equation to forecast the joint shear strength. Also, the experimental joint shear strength was compared with that predicted using the proposed framework model and the parametric equation. The results show that the Kernel regression predicted the shear strength of beam-column connections subjected to cyclic loading with the highest accuracy. Moreover, the squared R-value is 0.9752 which reflects the high efficiency of the Kernel model between other models. The results also reveal that the joint shear strength predicted using the Kernel regression is closer to the experimental values than the joint shear strength predicted using the parametric equation. As a result, the proposed model may be a useful tool for researchers and reinforced concrete engineers in accurately estimating the joint shear strength of beam-column connections (i) within the ranges of values used in this study for the input data, and (ii) with less time and cost than constructing other numerical schemes.