The interaction between the steel reinforcement and the surrounding concrete matrix through bond dominates the behavior of reinforced concrete structures. Despite its importance, bond behavior is rarely sufficiently addressed in previous studies and design guidelines, especially in extreme situations like fire exposure during construction. This study investigates the impact of elevated temperatures, reaching 825 °C, on bond strength via a series of pull-out tests performed on normal and high-strength concretes incorporating steel and polypropylene fibers. A machine learning model employing Decision Tree Regression ( \(DTR\) ) algorithms was created to forecast maximum bond strength at room temperature ( \(\tau_{b,20^\circ c}\) ). Hyperparameter tuning was conducted using two metaheuristic optimization techniques, Arctic Puffin ( \(AP\) ) Optimization and Energy Valley ( \(EV\) ) Optimization, to improve model accuracy and dependability. Additional procedures like as feature significance analysis, uncertainty quantification, and five-fold cross-validation were used to provide reliable models for the research. A dataset of 397 samples obtained from published publications was used, with 75% allocated for training and 25% for testing. The results demonstrate that the proposed machine learning framework serves as an effective and efficient instrument for predicting bond strength across diverse temperature settings. Given the provided information, it is very probable that both \(AP_{DTR}\) and \(EV_{DTR}\) will accurately calculate \(\tau_{b,20^\circ c}\) . With learning and assessing values of 0.0072 and 0.0097, respectively, and \(AP_{DTR} /EV_{DTR}\) ratios of 1.252 and 1.134, the \(EV_{DTR}\) yielded the lowest results in the \(MSLE\) measure. The \(AP_{DTR}\) reliability throughout the learning and evaluation phases, with values of 0.009 and 0.011, surpassed previous findings.
