Skip to content
Open access

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

2026 · IEEE Access · Vol 14, pp. 168308-168332 · 0 citations · 72 references

TL;DR

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.

Abstract

Most malware detection continues to rely on signature matching, which struggles with obfuscated or novel variants. This research presents a malware detector that instead relies on the static structure of function call graphs, summarized using L-moments of the degree, betweenness, closeness, and clustering measures into compact, interpretable descriptors that persist across malware variants. Evaluated on 175,633 binaries spanning many families under balanced, natural-prevalence, and prevalence-weighted training regimes, ensemble methods achieved Area Under the Curve (AUC) values up to 0.98 on the full-corpus random split, while a feed-forward neural network remained robust across regimes with AUC 0.92–0.95. Linear baselines were far more sensitive to training prevalence, with weighted $F_{1}$ dropping sharply under natural-prevalence training. Two leakage controls accompany the headline: a group-aware split preserving AUC 0.984, showing that duplicate leakage is not the primary driver of the result, and a collapse-to-unique bound of 0.915 on structurally novel samples. Ablation on a deduplicated 27-feature set shows that distributional moment summaries, rather than whole-graph statistics, carry most of the discriminative signal. 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. Explainability traces decisions to graph statistics using TreeSHAP on the production tree ensembles and to individual functions using a validated tree-model occlusion procedure.

Read PDF

Similar papers

Sep 2026

Partition-Matched Evaluation of Community Features under Distribution Shift in Android Malware Function-Call Graphs

Graph-based Android malware classifiers can lose accuracy under malware-type or family shifts. We test whether mesoscopic organization in function-call graphs provides shift-stable information beyond local degree profiles (LDP), global statistics, lightweight metadata, and size-matched random partitions. Using 15,000 M...

Jun-Ru Zhu, Yi-Xin Yang, Xiao-Qing Ding et al. · 0 citations
Open access 2026

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

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.

Zheng Lu, Thar Baker, De-Hua Zhu et al. · 0 citations
Open access Oct 2026

Classifier-guided discrete diffusion for malware family-conditional graph generation

We study whether discrete graph diffusion can be adapted to generate directed, weighted malware API call transition graphs in the low-data regime typical of dynamic malware analysis. Each execution trace is converted into a behavior graph whose nodes represent API types and whose directed edges represent transition cou...

Chandler Lu, Younghee Park, Fabio Di Troia · 0 citations
Review Open access Oct 2026

Revisiting Graph Representations for ML-Based Binary Code Similarity Detection: A Systematic Study

A systematic study of graph representations for ML-based BCSD is presented, finding that no single representation is universally dominant, that distinct topologies exhibit unique strengths depending on the evaluation scenario, and that their rankings are largely backbone-stable despite varying absolute performance.

Teng-Teng Yang, Yi-Kun Hu, Ji-Can Zhang et al. · 0 citations
#artificial intelligence Preprint Oct 2026

MARS: Malware Analysis with Rule-Based Scoring of LLM Claims

Large language models can triage malware through direct verdicts or behavioral claims scored by an external policy. We present MARS, a malware triage framework, and compare direct classification with single-pass claim scoring using the same evidence collector and identical static evidence bundles for each model. The ev...

H. Choi · 0 citations

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