The swift expansion of digital transactions has made financial fraud detection more crucial. Due to their incapacity to capture links between transactions, traditional machine learning models frequently fail to identify intricate and dynamic fraud patterns. Graph Neural Networks (GNNs) and traditional machine learning methods like Random Forest, Logistic Regression, and XGBoost are used in this paper to create a hybrid fraud detection system. By modeling transaction data as a graph with nodes representing entities and edges representing links between them, the suggested method enables the system to identify structural patterns and hidden dependencies. To improve detection performance, sophisticated GNN architectures such as Graph Convolutional Networks, Graph Attention Networks, and GraphSAGE are employed. According to experimental findings, graph-based techniques greatly increase the accuracy of fraud detection when compared to conventional methods. To enable real-time prediction and analysis, a Flask-based web application is used to further deploy the system. This study shows how well relational learning and feature-based models work together to detect fraud.