Temporal Analysis Framework for Intrusion Detection Systems: A Novel Taxonomy for Time-Aware Cybersecurity Journal: Computer Networks Note: Preprint submitted to Computer Networks
Tatiana S. Parlanti, Carlos A. Catania · 2025
Read Temporal Analysis Framework for Intrusion Detection Systems: A Novel Taxonomy for Time-Aware Cybersecurity Journal: Computer Networks Note: Preprint submitted to Computer Networks by Tatiana S. Parlanti, Carlos A. Catania (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-03799.txt.
Introduction
Tatiana S Parlanti and Carlos A Catania review how intrusion detection systems handle time, arguing that most still spot attacks only after serious damage is done. Their framework sorts network detection methods by how they treat time, from static checks on single flows to models that follow sequences across many windows.
Drawing on more than forty studies from 2020 to 2025, the pair find that sequential and window-based methods cover the widest range of attack stages, from reconnaissance through impact. Widely used datasets, they show, over-represent late-stage attacks and hold back early detection. The taxonomy is offered as groundwork for systems that anticipate threats rather than only react.
Catalog record
| CREATOR | Tatiana S. Parlanti, Carlos A. Catania |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 99 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-03799
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Temporal Analysis Framework for Intrusion Detection Systems: A Novel Taxonomy for Time-Aware Cybersecurity Journal: Computer Networks Note: Preprint submitted to Computer Networks Tatiana S. Parlanti, Carlos A. Catania Abstract Most intrusion detection systems still identify attacks only after significant damage has occurred, detecting late-stage tactics rather than early indicators of compromise. This paper introduces a temporal analysis framework and taxonomy for time-aware network intrusion detection. Through a systematic review of over 40 studies published between 2020 and 2025, we classify NIDS methods according to their treatment of time, from static per-flow analysis to multi-window sequential modeling. The proposed taxonomy reveals that inter-flow sequential and temporal window-based methods provide the broadest temporal coverage across MITRE ATT&CK tactics, enabling detection from Reconnaissance through Impact stages. Our analysis further exposes systematic bias in widely used datasets, which emphasize late-stage attacks and thus limit progress toward early detection. This framework provides essential groundwork for developing IDS capable of anticipating rather than merely reacting to cyber threats, advancing the field toward truly proactive defense mechanisms. 1 Introduction State-of-the-art Network Intrusion Detection Systems described in the literature face a temporal paradox: they excel at identifying attacks already in progress , i.e., when data is being exfiltrated, lateral movement concludes, or ransomware executes, corresponding to the Impact and other late-stage tactics in the MITRE ATT&CK framework [ 51 ] . Yet, they remain far less effective at capturing the subtle temporal patterns that characterize earlier adversarial behaviors, when detection would enable true prevention. This imbalance does not necessarily reflect poor model quality, but rather how intrusion detection has been conceptualized and assessed: systems are optimized to recognize clear, high-impact manifestations of compromise, rather than the gradual temporal evolution that precedes them. As noted by previous studies [ 34 , 92 , 69 ] , incorporating temporal information allows the detection of attacks that evolve over time or follow specific temporal distributions. Capturing the order and timing of events can reveal anomalies invisible in static analysis, since network traffic inherently forms a sequence of time-stamped interactions. Furthermore, by capturing how network behaviors evolve over time, temporal modeling also provides a foundation for assessing dynamic trust relationships among network entities. However, the field still lacks a coherent framework that explains how different temporal modeling strategies affect detection across the adversarial lifecycle. The central question, therefore, is whether explicitly modeling temporality can shift detection capabilities toward earlier attack stages. To address this question, we conduct an extensive survey and propose a temporal taxonomy that categorizes NIDS approaches according to their treatment of time—ranging from static per-flow models to sequential, window-based analysis of evolving network states. Using this taxonomy as an analytical framework, we examine over 40 recent publications (2020–2025) to assess current capabilities. Our analysis yields three key insights. First, the temporal modeling capabilities vary across existing approaches in the literature. However, to the best of our knowledge, there has been no unified framework to compare or classify them. As a result, there is no conclusive way to assess the current state of temporal detection in NIDS. The proposed taxonomy addresses this gap by defining five categories: Static Per-Flow, Static Contextual Snapshot, Intra-Flow Sequential, Inter-Flow Sequential, and Temporal Window-Based Sequential Analysis. Second, different temporal strategies align with specific stages of the MITRE ATT&CK lifecycle, offering distinct potential for intrusion detection. Third, widely used benchmark datasets tend to emphasize late-stage behaviors, constraining progress toward proactive intrusion detection. The major contributions of the paper are as follows. Novel Temporal Taxonomy. A five-category classification (S.1–T.3) that organizes time-aware NIDS approaches by their temporal analysis capabilities. ATT&CK Alignment. An analytical mapping linking the proposed taxonomy to the MITRE ATT&CK framework, clarifying the coverage of temporal approaches across more than 40 attack techniques. Dataset Bias Recognition. Identification of inherent biases in benchmark datasets (CIC-IDS2017, UNSW-NB15, NSL-KDD) that skew evaluation toward Impact-phase attacks. Systematic Review. A detailed analysis of recent work (2020–2025), highlighting strengths, limitations, and research directions for time-aware intrusion detection. The rest of this document is organized as follows. Section 2 exhibits the methodology used for the selection criteria of the papers presented in this survey. Section 3 presents fundamental concepts, then in Section 4 the related works are presented. In Section 5 a taxonomy is proposed for analyzing the literature, which is effectively carried out in Section 6 . The discussions and main limitations are then presented in Sections 7 and 8 , respectively. Finally, the conclusions are presented in Section 9 . 2 Methodology This systematic literature review examines the application of temporal and dynamic approaches in network intrusion detection systems, covering publications from 2020 through October 2025. The review process was designed to identify the most relevant studies while ensuring balanced coverage across the field. 2.1 Search Strategy and Information Resources The literature search was conducted across multiple complementary academic databases to capture diverse perspectives and publication venues in the field. The initial exploration was performed using the Crossref API [ 11 ] , chosen for its broad coverage across publishers and venues. The search was then expanded to include EBSCO [ 17 ] for additional journal coverage, Scopus [ 18 ] for its comprehensive indexing and citation analytics, IEEE Xplore [ 30 ] for publications in computer science and engineering, and the ACM Digital Library [ 5 ] for computing-focused research. 2.2 Query Construction and Filtering Process Search terms were organized around three conceptual pillars. First, terminology related to temporal characteristics, such as temporality, temporal, time-sensitive, time-series, dynamic . Second, terms that focus on the domain scope, like intrusion detection systems, IDS, NIDS . Finally, methodological approaches such as recurrent and sequential architectures, and graph-based methods were included. This conceptual structure allowed for capturing both traditional and emerging temporal approaches without restricting the search to a single technical paradigm. To ensure research quality and relevance, the selection was limited to peer-reviewed journal articles and conference proceedings published in Q1 or Q2 venues according to the Scimago Journal Rankings [ 80 ] . Each paper was required to explicitly address intrusion detection or network security and incorporate a temporal, dynamic, or sequential analytical component. We excluded works focused exclusively on domain-specific applications such as vehicular networks, medical monitoring, industrial control systems, or video surveillance, as well as preliminary or duplicate publications. 2.3 Selection and Analysis Process The selection process followed an iterative refinement cycle, evolving as insights emerged about how temporality is incorporated into intrusion detection research. Initial automated searches, particularly those via Crossref, yielded a broad pool of candidate studies; however manual inspection revealed that many addressed temporal aspects only marginally. Recognizing this, subsequent searches in EBSCO, Scopus, IEEE Xplore, and ACM Digital Library were refined and the methodology sections were manually examined to emphasize studies where temporal modeling played a central role. This iterative process, combining automated retrieval, conceptual refinement, and manual verification, resulted in a final corpus of 44 publications. The corpus includes both classical methods and recent innovations, providing a comprehensive foundation for understanding the current state and trajectory of the field. 3 Background An intrusion is any unauthorized activity that occurs within a network or system, which may or may not be malicious in intent [ 73 ] . When such activity aims to cause harm,