<p>This paper investigates the application of neural networks in predicting the power coefficient of wind turbines, with a focus on comparing ducted wind turbines to bare turbine systems. This study employed the actuator line method to obtain operational data across various scenarios. By designing innovative features for the neural network model, we successfully integrated data from ducted and bare turbines, enabling accurate power output predictions for diverse turbine configurations. The input data for the neural network model includes yaw angle, pitch angle, and blade tip speed ratio for ducted and bare wind turbines. The results demonstrate significant improvements in power output for ducted wind turbines, with the neural network effectively leveraging data from bare turbines to predict the performance of ducted systems. Specifically, the proposed model predicts ducted wind turbine power factors using bare turbine data with two distinct error regions: 1–3% at 0° pitch angle and 16–18% at − 5° pitch angle, with an average error of 10.01%. This outcome highlights the model’s adaptability while also revealing configuration-specific differences. However, the model exhibits limitations under certain operational conditions, particularly when the pitch angle deviates significantly at − 5°. Using data augmentation methods, the reported error for the pitch angle of − 5° was reduced to 14–16, while the error for the 0° pitch angle remained within the same range of 1–3. Consequently, the model’s average error decreased to 8.42. This research showcases the effective combination of computational methods and machine learning techniques to enhance wind turbine operational efficiency evaluation and forecasting.</p>