AppCloneGAN: A Generative Adversarial Network Approach for Detecting Fake Mobile Applications

The rise of counterfeit mobile applications, which imitate legitimate apps to trick users and spread malware, has created huge security and privacy issues. Static or dynamic code analysis is the basis of traditional detection methods, but it can be bypassed using code obfuscation and requires the app to be installed. This paper introduces AppCloneGAN, an innovative deep learning framework that makes use of Generative Adversarial Networks (GANs) to find fake applications based only on the analysis of their user interface (UI) screenshots. Our method addresses the detection problem as an image authenticity classification task. We incorporate a discriminator network within a conditional GAN framework that is trained to tell apart the screenshots of the original and copied applications that are malicious by recognizing in them such tiny visual detail like inconsistent branding, poor typography, misaligned elements, and low-quality graphical assets. The model has been trained and tested on a specially chosen dataset consisting of pairs of authentic and known malicious app UIs. The results of the experiments show that AppCloneGAN can accurately detect fake applications at a very high rate, thus becoming a promising, non-invasive, and scalable addition to the existing mobile security solutions. The system is able to function either as a pre-screening tool in app stores or as a client-side scanner to increase user protection.