Estimating the Torsional Capacity of Reinforced Concrete Beams Using ANFIS Models

Designing or appraising framed concrete buildings exposed to high eccentric loads requires precise reinforced concrete (𝑅𝐶) torsional strength estimates. Unfortunately, semi-empirical formulae still fail to adequately predict the torsional capacity of 𝑅𝐶 beams, particularly over-reinforced and strong ones. To overcome this limitation, accurate Machine Learning (𝑀𝐿) models might replace more sophisticated and computationally intensive models. This work evaluates and determines the most effective tree-based machine learning algorithms to estimate the torsional capacity (𝑇𝑟) of 𝑅𝐶 beams subjected to pure torsion. The objective of the present work is to provide innovative hybrid models that combine the concepts of the Adaptive neuro fuzzy inference systems (𝐴𝑁𝐹𝐼𝑆) model with other optimization approaches, such as the Giant trevally algorithm (𝐺𝑇𝐴) and Honey badger algorithm (𝐻𝐵𝐴), to accurately forecast the 𝑇𝑟. A total of 202 𝑅𝐶 rays were collected to form the data set. To facilitate the application of the 𝐴𝑁𝐹𝐼𝑆 models, a training set and a testing set were created from the database. Out of the 202 samples in the database, 25 percent (51) were used for evaluation and 75 percent (151) were used for learning. 𝐴𝑁𝐹 − 𝐺𝑇𝐴 demonstrated superior performance compared to 𝐴𝑁𝐹 − 𝐻𝐵𝐴, with a 50% higher performance in the learning phase and an 80% higher performance in the evaluation stage, as measured by the 𝑀𝑒𝑑𝑆𝐸 index values. The 𝐴𝑁𝐹 − 𝐺𝑇𝐴 obtained lower 𝑆𝑀𝐴𝑃𝐸 index values of 5.2192 and 5.39 compared to the values of 8.0622 and 9.3783 obtained by the 𝐴𝑁𝐹 − 𝐻𝐵𝐴 during the learning and assessment phases, respectively.