Musselwhite, Callum and Kruger, Bingle and Nitschke, Geoff (2026) Evolutionary Malware Data Augmentation, Proceedings of IEEE Congress on Evolutionary Computation (IEEE CEC 2026), 21-26 June, Maastricht, The Netherlands, IEEE Press.
![[thumbnail of 2026-Evolutionary Malware Data Augmentation.pdf]](https://pubs.cs.uct.ac.za/style/images/fileicons/text.png) |
Text
2026-Evolutionary Malware Data Augmentation.pdf Download (1MB) |
Abstract
Machine learning benefits malware detection and classification, yet scarce labelled data and rapidly evolving threats hinders progress. We investigate generative, evolutionary, and hybrid data augmentation in static malware detection given data scarcity and distributional shifts. We evaluate a generative adversarial network, evolutionary method, and hybrid method, with static features and Random Forest classifiers. Experiments across random scarcity, temporal holdout, and leave-one-familyout regimes (dataset treatments) indicate evolutionary augmentation provides consistent improvements if positive samples are extremely scarce and under temporal drift. Results also indicate the effects of leave-one-family-out evaluation are family dependent and as real data availability increases, augmentation benefits diminish, where training on real data alone becomes sufficient.
| Item Type: | Conference paper |
|---|---|
| Subjects: | Security and privacy > Intrusion/anomaly detection and malware mitigation Networks > Network properties > Network security |
| Date Deposited: | 15 Oct 2026 07:00 |
| Last Modified: | 15 Oct 2026 07:00 |
| URI: | https://pubs.cs.uct.ac.za/id/eprint/1820 |
Actions (login required)
 |
View Item |