This study develops and evaluates and evaluates a Global-Local Dynamic Directed Graph Neural Network for Parkinson’s disease detection using gait data. The researchers focused on vertical ground reaction force (VGRF) signals that were collected from pressure sensors embedded in wearable monitoring devices. These captured subtle motor differences between healthy patients and those with Parkinson’s disease. In order to address limitations in other models, this team designed a framework incorporating three modules: a Dynamic Graph Learning (DGL) unit to model evolving foot-pressure relationships, a Dynamic Directed Graph Network (DyDGN) to capture spatial features, and a Temporal Convolutional Network (TCN) to extract temporal patterns. The model was trained and validated on three data sets consisting of 93 patients with Parkinson’s and 73 controls. After evaluation, the Global-Local Dynamic Directed Graph Neural Network was performing better compared to the other models with an average improvement of 4.45% in accuracy. The authors emphasized that this model generalizes well across different datasets and effectively represents complex gait cycles.