Control flow graph neural networks vary in malware detection over time
Temporal Generalization and Explanation Stability of Control Flow Graph Neural Networks for Malware Detection
Cryptography and SecurityArtificial IntelligenceMachine Learning
Summary
Malware detection helps computers spot harmful software, but models tested only on past data might fail on newer malware. The authors studied how well graph-based neural networks, which analyze program structure, perform on malware from a later time. They found the choice of neural network design greatly affects how well the model handles new data. Surprisingly, simple models ignoring graph structure do best on past data but fail on future samples. This means testing malware detectors over time is crucial to find designs that remain accurate and explainable.
What this means in practice
- •For cybersecurity engineers: Select graph neural network designs that maintain malware detection accuracy on new software versions.
- •For software security teams: Improve evaluation protocols for malware detectors by testing their performance on future and shifted program samples.
Authors
Md. Asif Sajeed, Md. Nazrul Islam Mondal, Md Ashraful Hossen Akash
Abstract
Malware detection is a critical task in cybersecurity, and graph neural networks over control flow graphs have shown promising results for it. However, detectors are usually evaluated on a random split of a corpus collected over a single period, which cannot show how well a model generalizes to later samples. This study addresses that limitation with a strict temporal split: every model is trained on one period and scored once on a later one. Two corpora of control flow graphs, each node carrying 37 features, were extracted statically from 1,989 Windows portable executables: 459 graphs from 2024-2025 for training and 223 from 2026 for evaluation. Twelve variants and a flat-feature control were trained on the earlier corpus. The choice of message-passing operator changes robustness to the shift significantly, and every pairwise gap that survives correction separates an aggregating architecture from one built around a learned attentional readout. The ranking also reverses: the flat control, which sees node features but no topology, is the best in-distribution model and among the worst across the boundary, so a conventional benchmark would have rejected message passing. Neither recalibration nor ensembling substitutes for the operator choice. Attributions do not shift, but explanation validity is architecture-specific, and the most accurate operator on the later corpus is the hardest to explain. An architecture derived from the finding matches the best searched operator without search. The shift affects both malware and benign classes alike, so these are results about robustness to distribution shift, not malware evolution.