Load-carrying capacity and mode failure simulation of beam-column joint connection: Application of self-tuning machine learning model

The behavior of reinforced concrete external beam-column joint is highly stochastic and nonlinear due to the incorporation of several dimensional and concrete properties. Hence, establishing an accurate predictive for quantifying some beam-column joint characteristics is highly essential for structural engineering aspects. The current study is performed to predict load-carrying capacity (P-max) and mode failure of beam-column joint concrete using newly data intelligence model called extreme learning machine (ELM) model. 153 experimental data are gathered from the literature to construct the predictive model for training and testing phases. The input attributes consisted various dimensional information belong to the beam-column joint and concrete specification, are formed to be supplied for the predictive model. The proposed self-tuning predictive model validated against one of the prevalent regression model namely multivariate adaptive regression spline (MARS) model. The results evidenced that ELM model attained reliable prediction performance in comparison with MARS model. Statistical evaluation reported ELM and MARS models attained minimal root mean square error (RMSE approximate to 14.44 and 18.63), respectively. Accuracy of beam failure (BF) and joint failure (JF) predictions attained for ELM approximate to 0.78 and MARS approximate to 0.73. Overall, ELM model designated as a robust intelligence model can be developed for structural predesigned process and an alternative for empirical codes.