Self-Supervised Hypergraph–Causal Learning Framework for Unlabelled Threat Ecosystem Discovery in Passive DNS Streams

Authors

  • Banne Phanindhra Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation Deemed to be University
  • Kiran Kumar Chanumolu Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation Deemed to be University
  • Muni Nagamani G Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation Deemed to be University
  • Sri Hrasha Sanda Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation Deemed to be University
  • Moshe Dayan Sirapang Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation Deemed to be University
  • Anitha Rani Palakayala Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation Deemed to be University

DOI:

https://doi.org/10.19139/soic-2310-5070-3693

Keywords:

Passive DNS Analysis, Self-Supervised Cyber Threat Discovery, DNS Infrastructure Hypergraph Learning, Causal Threat Attribution, Unlabeled Threat Intelligence, Analysis

Abstract

The modern cyber threat landscape has been expanding its use of the dynamic aspects of internet naming systems through increasing amounts of passive DNS data; however, this creates many new malicious domains each day without an indicator as to whether they are known or unknown. This research addresses the necessity of autonomously finding such unknowns directly from unlabeled passive DNS streams. Most current detection frameworks rely upon large datasets and supervised classification approaches to identify malicious activity within a network, which limits their effectiveness in discovering new and emerging attack infrastructure and evolving domain ecosystems. Therefore, the proposed analytical framework will utilize self-supervised methodologies to model progressive levels of behavior, structure, time, and causality of DNS infrastructures. The proposed framework utilizes four primary modules to perform the above functions. These include: Temporal Entropy-Aligned DNS Representation Network (TEA-DRN), for learning embeddings based on behavioral patterns in DNS data; Hypergraph Structural Inference for Infrastructure Communities (HSIIC), for clustering communities of DNS infrastructure at higher order; Self-Supervised Threat Topology Evolution Model (STTEM), for analyzing the mutation of DNS infrastructures over time; Causal Infrastructure Attribution Engine (CIAE), for attributing orchestration dependency relationships among entities in DNS infrastructures; and finally, the Autonomous Unknown Threat Discovery Model (AUTDM), for discovering unknown malicious activity at the ecosystem level. Results from experimental testing demonstrate increased discovery coverage of previously unknown malicious infrastructures that can lead to earlier detection of coordinated threat networks and enhance the proactive cyber defense intelligence capabilities of organizations.

Downloads

Published

2026-11-16

How to Cite

Banne Phanindhra, Chanumolu, K. K., G, M. N., Sanda, S. H., Sirapang, M. D., & Palakayala, A. R. (2026). Self-Supervised Hypergraph–Causal Learning Framework for Unlabelled Threat Ecosystem Discovery in Passive DNS Streams. Statistics, Optimization & Information Computing. https://doi.org/10.19139/soic-2310-5070-3693

Issue

Section

Research Articles