Droplet-based microfluidics is a transformative technology with applications across diverse scientific and industrial domains. However, predicting the droplet size generated by individual microchannels before experiments or simulations remains a significant challenge. In this study, we focus on a double T-junction microfluidic geometry and employ a hybrid modeling approach that combines machine learning with metaheuristic optimization to address this issue. Specifically, particle swarm optimization (PSO) is used to optimize the hyperparameters of a decision tree (DT) model, and its performance is compared with that of a DT optimized through grid search (GS). The hybrid models are developed to estimate the droplet diameter based on four parameters: the main width, side width, thickness, and flow rate ratio. The dataset of more than 300 cases, generated by a three-dimensional numerical model of the double T-junction, is used for training and testing. Multiple evaluation metrics confirm the predictive accuracy of the models. The results demonstrate that the proposed DT-PSO model achieves higher accuracy, with a coefficient of determination of 0.902 on the test data, while simultaneously reducing prediction time. This methodology holds the potential to minimize design iterations and accelerate the integration of microfluidic technology into the biological sciences.