Random Forest over Semantic PDF Object Graphs for MalwareDetection
IJAICET - International Journal of Artificial Intelligence, Cybersecurity and Emerging Technologies · 2026
DOI: 10.68307/e7agqk07 ↗Kurz gefasst
Fachartikel in der Zeitschrift IJAICET - International Journal of Artificial Intelligence, Cybersecurity and Emerging Technologies, 2026. Verfasst von Mahesh Kumar Bagwani. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Portable Document Format (PDF) malware remains a practical attack vector because document objects, embedded actions,streams, and parser-dependent structures can conceal malicious behaviour. This paper presents an IJAICET-formatted empiricalstudy of Random Forest classification over semantic PDF object-graph embeddings. Each PDF is represented as an Object ReferenceGraph and its BERT-65k PDFObj2Vec node embeddings are transformed into a fixed-length feature vector using mean, maximum,minimum, and standard-deviation pooling together with graph-level structural statistics. A 200-tree Random Forest classifier istrained on the public PDFObj2Vec evaluation dataset. The proposed baseline achieves 99.97% accuracy on the baseline test split and96.10% accuracy on the extended split, close to the reported BERT-65k+GIN extended accuracy of 96.62%. The results show thatsemantic object embeddings retain strong discriminative evidence even when graph message passing is replaced by an interpretableclassical ensemble model.
Zitieren
APA
Bagwani, M. K. (2026). Random Forest over Semantic PDF Object Graphs for MalwareDetection. IJAICET - International Journal of Artificial Intelligence, Cybersecurity and Emerging Technologies. https://doi.org/10.68307/e7agqk07
Vancouver
Bagwani MK. Random Forest over Semantic PDF Object Graphs for MalwareDetection. IJAICET - International Journal of Artificial Intelligence, Cybersecurity and Emerging Technologies. 2026. doi:10.68307/e7agqk07
BibTeX
@article{Bagwani2026,
author = {Bagwani, Mahesh Kumar},
title = {Random Forest over Semantic PDF Object Graphs for MalwareDetection},
journal = {IJAICET - International Journal of Artificial Intelligence, Cybersecurity and Emerging Technologies},
year = {2026},
doi = {10.68307/e7agqk07}
}
EUVIMED