Improving Detection Accuracy for Malicious JavaScript Using GAN

摘要

Dynamic web pages are widely used in web applications to provide better user experience. Meanwhile, web applications have become a primary target in cybercriminals by injecting malware, especially JavaScript, to perform malicious activities through impersonation. Thus, in order to protect users from attacks, it is necessary to detect those malicious codes before they are executed. Since the types of malicious codes increase quickly, it is difficult for the traditional static and dynamic approaches to detect new style of malicious code. In recent years, machine learning has been used in malicious code identification approaches. However, a large number of labeled samples are required to achieve good performance, which is difficult to acquire. This paper proposes an efficient method for improving the classifiers' recognition rate in detecting malicious JavaScript based on Generative Adversarial Networks (GAN). The output from the GAN is used to train classifiers. Experimental results show that our method can achieve better accuracy with a limited set of labeled sample.