This paper introduces ASTRID, a GAN-based model that can identify malware using API call sequence analysis in noisy settings, a typical problem in disk forensics. The new framework addresses the shortcomings of conventional detection approaches by leveraging adversarial learning to differentiate between benign and malicious sequences. API call information from two noisy datasets was utilized to train the model under real-world-like conditions, while a third, unseen dataset was employed to test the model's generalization ability. ASTRID achieved an accuracy of 96.8% on training data and 95.5% on the test dataset, with performance comparable to state-of-the-art models. These outcomes demonstrate the power of ASTRID to tackle noise and provide detection reliability, providing an impressive solution towards durable malware detection through sequence-based learning.
