The ability to predict flight prices accurately is essential for airlines and travelers, yet the complex air transportation network and dynamic pricing pose significant challenges. This study develops a hybrid model that combines a Graph Attention Network (GAT) with standard flight and time features, illustrating the interconnections in networks and variations among flights. A comparative analysis assesses its performance against traditional machine learning methods, including Random Forest, XGBoost, and Linear Regression, highlighting the advantages of ensemble learning. The hybrid GNN ensemble achieves strong results on a Bangladeshi flight dataset, with a Mean Absolute Error (MAE) of 31724.48 BDT and a Root Mean Square Error (RMSE) of 54586.57 BDT, comparable to or surpassing classical methods. Additionally, interpretability analysis using the SHAP technique reveals key factors influencing prices, offering insights into the model’s decision-making. This study enhances airfare prediction by effectively incorporating network structure and complex elements, benefiting both consumers and the airline industry