This research presents a novel approach of artificial intelligence (AI) based gene expression programming (GEP) for predicting the lateral load carrying capacity of RC rectangular columns when subjected to earthquake loading. To achieve the desired research objective, an experimental database assembled by the Pacific Earthquake Engineering Research (PEER) center consisting of 250 cyclic tested samples of RC rectangular columns was employed. Seven input variables of these column samples were utilized to develop the coveted analytical models against the established capacity outputs. The selection of these input variables was based on the linear regression and cosine amplitude method. Based on the GEP modelling results, two analytical models were proposed for computing the flexural and shear capacity of RC rectangular columns. The performance of both these models was evaluated based on the four key fitness indicators, i.e., coefficient of determination (R²), root mean squared error (RMSE), mean absolute error (MAE), and root relative squared error (RRSE). From the performance evaluation results of these models, R², RMSE, MAE, and RRSE were found to be 0.96, 53.41, 38.12, and 0.20, respectively, for the flexural capacity model, and 0.95, 39.47, 28.77, and 0.22, respectively, for the shear capacity model. In addition to these fitness criteria, the performance of the proposed models was also assessed by making a comparison with the American design code of concrete structures ACI 318-19. The ACI model reported R², RMSE, MAE, and RRSE to be 0.88, 101.86, 51.74, and 0.39, respectively, for flexural capacity, and 0.87, 238.74, 183.66, and 1.35, respectively, for shear capacity outputs. The comparison depicted a better performance and higher accuracy of the proposed models as compared to that of ACI 318-19.

This chapter explores an AI-powered approach for predicting the seismic capacity of RC rectangular columns using multi-expression programming (MEP). Leveraging a database of 250 RC column specimens tested under seismic loads from the Pacific Earthquake Engineering Research (PEER) Centre, two distinct MEP models were developed for flexural and shear capacity prediction. These models, trained with five key input variables using MEPX software, achieved high accuracy (R2 > 0.96) exceeding the performance of ACI 318-19 code provisions. Additionally, the models captured the underlying physical processes of seismic behavior in columns. These findings suggest the potential of MEP-based models for practical application in seismic design automation of RC structures. Request access from your librarian to read this chapter's full text.