<p>Accurately predicting the torsional capacity of reinforced concrete (RC) beams may be crucial for the design or evaluation of framed concrete buildings under high eccentric loadings. Regretfully, the torsional capacity of RC beams, particularly those that are over-reinforced and of high strength, cannot yet be reliably predicted using conventional semi-empirical models. By creating precise Machine Learning (ML)-based models as an alternative to other more intricate and computationally intensive models, this limitation may be overcome. The objective of this work is to assess and identify the most effective ML algorithms based on trees for predicting the torsional capacity (T_r) of RC beams subjected to pure torsion using dataset collected from literature containing 202 samples. In order to accomplish this objective, two tree-based methodologies, known as Random Forests (RF) analysis, applied on the computer. Two metaheuristic optimization algorithms named Artificial Hummingbird Optimization (AHO) and Beluga Whale Optimization (BeWO) linked to RF analysis to gain the determinative parameter's proper values. Prediction algorithms developed using three input factors: the reinforcing factor, the concrete's compressive strength, and the cross-sectional area. There is a great potential for AHO-RF and BeWO-RF to provide precise estimates of the concrete's T_r of RC.&#xa0;It is evident from computed statistical measures for performance analysis BeWO-RF outperforms AHO-RF and other models.</p>