The grooving method has shown significant potential in addressing premature debonding in externally bonded fiber-reinforced polymer (FRP) composites on concrete members in experimental studies. However, accurately predicting the bond strength between FRP and grooved concrete remains a challenging task due to the complexity of the bond behavior and the limited availability of comprehensive experimental data. This study explored the use of machine learning (ML) models to address these challenges, focusing on enhanced prediction of bond strength through synthetic data-driven deep learning. Two ML models, Dense Neural Network (DNN) and Random Forest (RF), were initially applied to predict bond strength. To overcome data limitations, the Gaussian Copula Synthesizer (GCS) was employed to generate 5,838 synthetic data points, enriching the training dataset. The enhanced DNN model trained on the expanded dataset demonstrated significant performance improvements, with reductions of 24% and 25% in RMSE and MAE, respectively, and a 10% increase in R2. These findings emphasize the value of incorporating synthetic data to enhance the accuracy and reliability of ML-based predictive models for the bond strength of FRP-to-grooved concrete. This study provides a novel model incorporating synthetic data generated by GCS to enhance machine learning predictions, offering a more accurate and reliable model for predicting the bond strength of FRP-to-grooved concrete joint.