In recent years, fiber-reinforced polymers (FRP) have emerged as a highly effective solution for strengthening reinforced concrete (RC) structures. However, accurately assessing the fire resistance of FRP-strengthened members remains a significant challenge due to the limited guidance available in current building codes, often leading to conservative and cost-intensive evaluations. Experimental testing and numerical analysis required for such assessments are resource-demanding, highlighting the need for more efficient methods. This study investigates the application of machine learning (ML) techniques to predict the fire resistance of FRP-strengthened RC beams. Twelve ML models, including eight ensemble methods and four traditional approaches, were employed. The models were trained using a comprehensive dataset comprising over 21,000 data points obtained from numerical simulations and experimental tests. The dataset captured variations in geometric configurations, insulation strategies, loading conditions, and material properties. To enhance predictive accuracy, Bayesian optimization and k-fold cross-validation were applied for model tuning, while the Shapley Additive Explanations (SHAP) method was utilized to assess the relative importance of features influencing fire resistance. Among the models tested, Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Light Gradient Boosting (LGBoost), and Gradient Boosting (GRB) demonstrated superior performance, achieving accuracy rates exceeding 92%. Key factors identified as significantly affecting fire resistance included loading ratio, area of tensile reinforcement, insulation depth, concrete cover thickness, and FRP area. The findings underscore the potential of ensemble ML techniques over traditional methods for accurately predicting the fire resistance of FRP-strengthened RC beams, offering critical insights for optimizing design practices and enhancing structural fire safety.