A GAN-Based Approach for Identifying Fake Accounts on Twitter

Abstract

The multiple security threats on the network make the need for robust security measures a major concern. The increasing presence of fake accounts and malicious actors on online platforms poses significant challenges, requiring sophisticated detection techniques to maintain network integrity. To address these issues, we propose a novel method for detecting fake accounts by leveraging Generative Adversarial Networks (GANs). By analyzing data extracted from platform APIs, our approach leverages the unique characteristics of GANs to improve the accuracy and efficiency of the detection process. In this study, we develop a GANs-based model specifically designed to detect fake accounts. The model is built through several key stages: first, we collect a comprehensive dataset, then perform data processing and preprocessing to make it suitable for machine learning applications. Next, the model is trained using various hyperparameters to optimize accuracy, thus learning the underlying patterns associated with fake accounts. After the training stage, the model is tested on previously unseen data to evaluate its generalization and performance in real-world scenarios. Experimental results show that our model achieves a threshold value of 0.0054779826. This value plays a crucial role in determining the accuracy of the detection system. The smaller the threshold value, the higher the model accuracy, as it shows a lower error rate in distinguishing between real and fake accounts. The ability of GANs-based models to adaptively learn from data during the training process contributes to high precision in detecting anomalies as well as minimizing false positives.