Abstract: Atrial fibrillation is a common cardiovascular disease with an increasing incidence worldwide. The hemodynamic patterns within the left atrium and left atrial appendage in atrial fibrillation patients, such as velocity and pressure fields, are important indicators for assessing the risk of thrombus formation. The conventional research method for cardiovascular issues is computational fluid dynamics, which, although recognized for its accuracy, is limited in its clinical application due to high computational costs. To address this issue, machine learning methods have been introduced, proposing a neural network architecture based on edge convolutional layers and dual channels. After training, this architecture can predict the internal velocity and pressure fields from a point cloud model of the left atrial appendage. The architecture achieves prediction errors of around 10% for velocity fields and 5% for pressure fields on the test dataset. Hemodynamic analysis reveals that atrial fibrillation patients have lower flow velocities in the left atrial appendage and higher average pressures in the left atrium compared to healthy individuals, and males are more prone to atrial fibrillation than females.