Accurate diagnosis of mental health disorders like schizophrenia, anxiety, and mood disorders is often hindered by insufficient consultation time and rushed clinical evaluations, leading to inconsistent results and delayed treatment. Electroencephalography (EEG), a completely non-invasive method for recording electrical brain activity, it’s also cost-effective. However, conventional analysis techniques struggle to interpret the complex patterns present in EEG signals effectively. In this research work, a graph-based methodology is introduced where EEG recordings are converted into structured graph representations. Graph Neural Networks (GNNs) are employed because of their exceptional capacity to process and evaluate intricate relationships within data. Each node of the graph contains detailed information about brain wave characteristics. Additional personal information, including the participant’s age, gender, education, and intelligence quotient score, is included to provide more comprehensive data for analysis. The study specifically implements a Graph Isomorphism Network (GIN), an advanced type of GNN, as the main analytical framework. Its effectiveness in differentiating healthy persons from those dealing with psychiatric conditions is systematically compared against several well-known machine learning approaches. Across multiple evaluations comparing healthy controls versus specific psychiatric disorders, the GIN model shows superior performance in every case.