Investigating shear capacity prediction in concrete deep beams using data-driven techniques

Prediction of the shear strength of reinforced concrete deep beams (RCDBs) is essential to guarantee the structural integrity and functionality of the designed structure. Despite continuous efforts to improve design guidelines and standards, it is still very difficult to estimate the shear capacity of RCDBs accurately. In this research, most challenges were overcome by adopting a full dataset of 503 shear strength measurements for RCDBs, with and without shear reinforcement. Machine learning and soft computing data-driven techniques, such as gene expression programming (GEP) and multi-objective evolutionary polynomial regression (MOGA-EPR), were used to develop predictive models. The results were most impressive, with coefficients of determination (R2) of 0.83–0.95 for RCDBs with shear reinforcement and 0.83–0.90 for those without. Comparatively, the traditional design code and previous studies report much smaller R2 values, in the range 0.32–0.67 for RCDBs with shear reinforcement and 0.16–0.18 for those without. The main outcome of this research is the construction of two separate predictive models through GEP and MOGA-EPR, which will help engineers estimate shear strength more reliably in the case of deep beams with and without shear reinforcement.