Connections are crucial zones in steel buildings since they provide interaction between principal structural components (i.e., beams, columns) and provide stability to the entire building. Therefore, predicting the connection behavior remains essential to achieving safe structures. This paper proposes novel hybrid ANN-HPO models, which are made by integrating artificial neural networks (ANN) and hunter-prey optimization (HPO), for predicting the moment-rotation (M-θ) behavior of steel flush endplate (FEP) connections. For this purpose, a database of 121 steel FEP connections and corresponding data points of the M-θ curve is compiled to develop the ANN-HPO models. The geometric and material properties of the column, beam, endplate, and bolts are used as input variables. Meanwhile, the M-θ behavior, represented by the ultimate moment (M_u), the rotation at the ultimate moment (θ_u), the initial stiffness (R_ki), the reference plastic rotation (θ₀), and the shape parameter (n) are output variables. To assess the predictive power of the ANN-HPO model, the obtained results are compared with those of four recently robust machine learning (ML) models, including decision tree (DT), random forest (RF), gradient boosting regression tree (GBRT), and extreme gradient boosting (XGBoost). The comparison shows that the proposed ANN-HPO model has the best performance for predicting the M-θ behavior in the training and testing phases. Finally, a graphical user interface (GUI) application is developed to enable users to predict the M-θ behavior and reconstruct M-θ curves of steel FEP connections based on trained ANN-HPO models.