<p>The rapid proliferation of the internet and online e-commerce platforms has made it easier for users to access the products and information they need, while also significantly increasing the volume of information available. In this context, recommendation systems play a crucial role by providing personalized recommendations based on user preferences. However, the open and user-interaction-based nature of these systems allows fake users to interfere with the system and skew recommendation results. Fake users create bias within the system to highlight or discredit specific products on online e-commerce platforms, negatively impacting both recommendation quality and user trust. Therefore, the accurate and effective detection of fake users has become a critical research problem. This study presents an approach to detecting fake users based on user’s behavioral interaction patterns. The proposed method extracts behavioral features such as rating frequency, temporal activity and the consistency of review behavior. Subsequently, a Graph Attention Network (GAT) is used to model the relationships between users and products. The SMOTE and ADASYN methods are applied to address the class imbalance problem. Finally, classification is performed using machine learning algorithms. Experimental results demonstrate that the proposed approach achieves high success in detecting fake users. In particular, when behavioral features were integrated with GAT alone, the highest F1 score of 99.10% was achieved by the Logistic Regression model. The findings demonstrate that the proposed method offers an effective and practical solution for enhancing the reliability of recommendation systems in online e-commerce environments.</p>