PurposeThe purpose of this study is to perform a numerical analysis of fiber-reinforced polymer (FRP) retrofitting in reinforced concrete (RC) joints at high temperatures and predict models using artificial neural networks (ANN). The aim was to gain insights into their structural behavior across a range of loading conditions from room temperature to 800°C. Additionally, the research assessed the efficiency of carbon fiber-reinforced polymer (CFRP), glass fiber reinforced polymer (GFRP) and aramid fiber reinforced polymer (AFRP) strengthening in enhancing the structural performance of the critical sections.Design/methodology/approachThe linear numerical simulations were conducted to evaluate the performance of RC beam-column joints using finite element modelling (FEM) analysis. The ANN model demonstrated exceptional effectiveness in predicting the stiffness of frames with openings, establishing itself as the premier machine learning algorithm for frame stiffness estimation. In the conventional model, 300°C was proven to be an effective temperature approach. Subsequently, maintaining a constant temperature of 300°C, an in-depth analysis of nearly 30 models of three retrofitting techniques was conducted under thermomechanical loading.FindingsThe CFRP retrofits yielded 15% less deflection and 30% more stress than the remaining FRPs, and the ANN models predicted the deflection, main stresses, bending moment and shear force. The ANN model results were compared with those of other frequently used models. The R thresholds (R = 0.954, 0.981, 0.986, 0.968, 0.978 and 0.936) for training, testing and validation indicated that the ANN model achieved data variability. The findings indicate that the ANN model is more accurate because of the strong connection between the numerical model and the prediction.Originality/valueTo identify the pinpoint of critical segments within the rehabilitation section and determine the most effective wrapping method among the three laminates.