On the Efficiency of Advanced Machine Learning and Deep Learning Regression Techniques for Predicting the Punching Shear Strength of Reinforced Concrete

Punching shear is a complicated dilemma that is affected by a combination of many variables and several mechanisms. The aim of this research is to investigate the application of regression analysis methods in data analysis and prediction. Regression analysis is a powerful statistical technique that enables researchers to model the relationship between a dependent variable and one or more independent variables. In this study, we explore various regression analysis methods, including multiple machine learning and deep learning-based regression algorithms, to predict the punching shear strength of reinforced concrete. We compared the results from the recent state-of-the-art models and our models, and we concluded that the best models for predicting the punching shear were as follows: (1) unsupervised Deep learning Autoencoder with R ² score of 0.91; (2) Extreme Gradient Boost Regression with R ² score of 0.8959; (3) Decision Tree Regression with R ² score of 0.9046; (4) Genetic Algorithm Optimization Random Forest Regression with R ² score of 0.9015. After applying the grid search to search for the best hyperparameters to get the best fit for the data and adding it to the genetic algorithm optimization model with the random forest regression model, we found a huge improvement in the prediction accuracy, as the random forest regression model R ² score improved from 0.8951 to 0.9015 and with an RMSE from 92.52 to 89.65. The worst models were long short-term memory (LSTM) and support vector regression (SVR) models, both of them showed substantially inferior predictive ability in comparison to other models with R ² -0.8282 and 0.7465, respectively.