Pulmonary hypertension (PH) is characterized by a significant increase in mean pulmonary arterial pressure at rest and can initiate substantial right ventricular remodeling, often leading to right heart failure. The diagnosis of PH primarily relies on assessing the hemodynamics of the pulmonary circulation. While non-invasive modalities such as four-dimensional flow MRI and Doppler imaging assist with the diagnosis, PH is confirmed only through invasive right heart catheterization. The challenges of measuring arterial pressures non-invasively have prompted the use of computational fluid dynamics (CFD) models and simulation to assess pulmonary hemodynamics in silico. However, such CFD models have not been widely adopted in clinical practice due to high computational costs and complex deployment procedures. In this study, we introduce a surrogate machine learning (ML) model that circumvents complex CFD methods to predict pulmonary artery differential pressures. A multi-layer feed-forward neural network (MFNN) was trained using the pulmonary branch inlet velocity, alongside pulmonary vascular resistance modeled by the three-element Windkessel formula. The ML model demonstrated remarkable predictive accuracy with an R2 of 92%. Integrating such a pipeline directly with clinical imaging modalities promises to provide a non-invasive and feasible alternative for assessing pulmonary pressure gradients.