In this study, a data-driven framework for parametric analysis, machine learning-based alternative modeling, and multi-objective optimization of a fluid propulsion system based on the Coanda effect for UAV applications is presented. A database of 1890 design points was generated using ANSYS Fluent software and the k-ε realizable turbulence model. Six machine learning algorithms were trained using ten physically engineered features to predict propulsion performance indices. The XGBoost model with a coefficient of determination of 0.9928 for the discharge ratio and the Multilayer Perceptron (MLP) model with a coefficient of determination of 0.9968 for the thrust showed the best performance. Gini sensitivity analysis showed that the inner radius with a 72% contribution has the greatest impact on the discharge ratio, while the inlet mass flow rate with a 30% contribution plays an important role in the thrust generation. Multi-objective optimization using the NSGA-III algorithm extracted the Pareto front between flow enhancement and thrust, and configurations with thrust up to 32 N at a mass flow rate of 100 g/s were identified. The results showed that the optimized propulsion system is able to provide 87% of the equivalent convergent-divergent nozzle thrust with a 25 dB reduction in noise level and a 45% reduction in exit velocity. These results demonstrate the potential of fluid propulsion systems based on the Coanda effect for the development of low-noise propulsion systems suitable for UAV applications.