Skip to content
Open access

Learning Higher-Order Dependencies in Malware Control-Flow Graphs With Hypergraph-Level Explanations

2026 · IEEE Transactions on Information Forensics and Security · Vol 21, pp. 9467-9481 · 0 citations · 58 references

TL;DR

Results indicate that the proposed MalHGNN approach improves malware classification performance while providing interpretable node- and hyperedge-level evidence for its predictions, demonstrating the benefits of higher-order dependency modeling.

Abstract

Malware continues to cause significant security and financial losses, creating a pressing need for accurate and scalable detection and analysis. Accordingly, static malware detection has become a widely used and scalable first line of defense.However, reliable static detection remains challenging because discriminativemalware behavior can be distributed across distant and non-adjacent regions of a program. Many malware samples exhibit staged execution workflows, so the corresponding evidence may span multiple basic blocks that are not directly connected in the control-flow graph (CFG). Conventional graph-based detectors propagate information only through local control-flow links and struggle to capture such long-range dependencies. Moreover, while graph-based detectors offer a more semantically grounded view of program behavior, few works provide actionable explanations for their predictions. To address these issues, we propose MalHGNN, a malware classification framework that elevates the CFG into a hyper control-flow graph (HCFG) for higher-order dependency modeling. MalHGNN introduces explicit higher-order shortcuts by combining semantic k-NN hyperedges with structural k-Hop hyperedges, enabling aggregation of features across distant, non-adjacent basic blocks. In addition, we develop an interpretability technique, HGNN-CAM. It provides both node-level and hyperedge-level attributions to highlight critical basic blocks and their higher-order combinations. Extensive experiments on the BIG2015 and MalwareBazaar real-world datasets validate the effectiveness of MalHGNN. MalHGNN consistently outperforms representative feature-based and graph-based malware detectors on malware family classification, demonstrating the benefits of higher-order dependency modeling. These results indicate that our approach improves malware classification performance while providing interpretable node- and hyperedge-level evidence for its predictions.

Read PDF

Similar papers

Sep 2026

Behavior-Centric Malware Classification with Fine-Grained Malicious Logic Localization

Effective malware analysis requires understanding not only whether a program is malicious, but also which behaviors it exhibits and where those behaviors originate in the code. Existing machine-learning-based malware detectors largely operate as black boxes, providing limited insight into the malicious logic responsibl...

P. Baral, Zhuo-Yun Qian, Hai-Lu Xu et al. · 0 citations
Open access 2026

Graph-Theoretic Analysis of Function Call Graphs Using L-Moments for Malware Detection

Family-wise cross-validation confirms performance holds on unseen malware families, and leave-one-source-out and source-prediction probes show that detection holds within a single source, although source and class remain confounded in the pooled corpus.

Anthony J. Rose, Wayne C. Henry, Scott R. Graham et al. · 0 citations
Sep 2026

FiCoVuL: A Framework for Fine-grained and Cross-function Code Vulnerability Detection

FiCoVuL is presented, a framework for analyzing interconnected functions and providing fine-grained guidance for vulnerability fixing that significantly outperforms other methods in both vulnerability detection and localization.

Hong-Jun Huang, Fu-Tai Zou, Jia-Ping Gui et al. · 0 citations
Sep 2026

Stress-Testing Structure-Aware Calibration of Malware Graph Neural Networks under Type Shift

Post-hoc malware calibrators can condition confidence on graph structure, but their structural inputs may leave the support represented by validation data under malware-type shift. We study this risk on MalNet-Tiny by holding out each of four malware types across three seeds, freezing a graph isomorphism network, and f...

Jun-Ru Zhu, Yi-Xin Yang, Xiao-Qing Ding et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.