The DOI matches its Crossref record; 94 of the 94 references it cites were found in a registry or an index; 11 of the 70 sources its claims cite were checked for support.
A Review of Recent Advances, Challenges, and Opportunities in Malicious Insider Threat Detection Using Machine Learning Methods
Open accessAbstract
Insider threat detection has become a paramount concern in modern times where organizations strive to safeguard their sensitive information and critical assets from malicious actions by individuals with privileged access. This survey paper provides a comprehensive overview of insider threat detection, highlighting its significance in the current landscape of cybersecurity.
Overview
CoChat Summary
TL;DR
This is a survey of malicious insider threat detection using machine learning, deep learning, and NLP, with emphasis on recent methods, datasets, metrics, challenges, and future directions. It argues that learning-based approaches are promising but still limited by data scarcity, interpretability, class imbalance, and real-world deployment constraints.
Key findings
- The paper states that insider threats were responsible for 35% of data breaches in the 2023 Verizon DBIR, that malicious insider threats can inflict 20 times greater damage than external actors, and that the average cost is $4.5 million per breach.
- It identifies four major categories of insider threats: malicious, accidental, fraudulent, and privileged vs. non-privileged, and also discusses duration as another categorization dimension.
- The review says machine learning and deep learning methods have shown promise for insider threat detection, especially behavior-based, anomaly-based, and graph-based approaches; it also highlights the growing use of RNNs, LSTMs, CNNs, GNNs, autoencoders, and NLP.
- The paper notes recurring practical issues: false positives, false negatives, class imbalance, need for large labeled datasets, interpretability problems, and adversarial attacks.
Methods
- This is a narrative review/survey paper rather than an empirical study; it synthesizes prior work on insider threat detection.
- It reviews and compares prior studies, datasets, techniques, and metrics, including examples such as CERT, TWOS, RUU, and Sentiment140, and summarizes methods in Tables 1–4.
- The paper’s main analysis is a structured taxonomy covering threat types, detection approaches, data sources/features, learning-based techniques, evaluation metrics, challenges, and future trends.
Limitations
- The paper states that the field is limited by data scarcity, model interpretability, and adversarial attacks, and calls for explainable AI, hybrid models, and more benchmark datasets.
- It also states that many methods depend on large labeled datasets or otherwise struggle with false positives and class imbalance.
- Evident from the text: because this is a survey, it does not report a new experiment or unified benchmark result of its own; instead it compiles and summarizes results from prior studies.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
6 Valid · 6 Not valid · 40 Not checked · 47 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Unsupervised learning approaches, including clustering algorithms like K-Means and hierarchical clustering, are adept at grouping similar behavioral patterns and flagging outliers [61], [70], [71], [72] .”
What the sources say
- Not checked — no full text of the sourcecites [61] source
- Not checked — no full text of the sourcecites [70] source
- Partly supportedcites [71] Noever, 2019, Classifier suites for insider threat detection, arXiv:1901.10948 sourceExperimental — accuracy not yet measured
The Discussion and Conclusions says unsupervised methods can cluster individuals’ network traffic by activity type and may help identify previously unseen threats. It discusses k-nearest neighbors and auto-encoders, not K-Means or hierarchical clustering, and does not demonstrate that those named algorithms flag outliers. (LLM, conf=0.86)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Rule-based detection is effective in identifying known attack patterns or policy violations, offering control over the detection process for customization to specific security requirements [64], [66].”
What the sources say
- Could not verifycites [64] sourceExperimental — accuracy not yet measured
The Related Work section describes matching command sequences and using machine-learning methods to detect malicious behavior, but it does not characterize rule-based detection or assess its effectiveness against known attack patterns or policy violations. Searches for rule-based detection, predefined patterns, and policy violations found no passage addressing the attributed claim; page numbers were not available in our copy. (LLM, conf=0.72)
- Could not verifycites [66] sourceExperimental — accuracy not yet measured
Our indexed copy of Elmrabit et al. (2020) contains only repository/metadata material — the title and author block (unnumbered cover page), the Scopus/RIS and BIBTEX citation records (Abstract, pp. n.), the licence and embargo note (Abstract), and the research-fingerprint/subject-area listing (Abstract). No body text (introduction, related work, or methodology discussion) is retrievable, so nothing in our copy speaks to rule-based detection, known attack patterns, policy violations, or control/customization for security requirements. Searched queries: 'rule-based detection known attack patterns policy violations', 'signature based intrusion detection limitations insider threat customization security requirements', 'introduction insider threat detection methods anomaly based machine learning', 'related work rule based approach known patterns advantages disadvantages' — all returned only the metadata passages. Absence here is not evidence against the claim; the attribution cannot be checked on our copy. (LLM, conf=0.72)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Additionally, autoencoders, a form of unsupervised deep learning, have shown promise in reconstructing normal patterns and identifying anomalies [78].”
What the sources say
- Could not verifycites [78] sourceExperimental — accuracy not yet measured
The paper’s sensor-extraction section describes detecting deviations from normal activity, but does not discuss autoencoders or reconstructing normal patterns. Searches for autoencoders and reconstruction-based anomaly detection found no passage addressing those methods. (LLM, conf=0.75)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Several studies [45], [49], [55], [57] , [62] have demonstrated that deep learning, characterized by the use of neural networks with multiple layers, has remarkable capabilities in capturing intricate patterns within complex datasets.”
What the sources say
- Not checked — no full text of the sourcecites [45] source
- Partly supportedcites [49] sourceExperimental — accuracy not yet measured
The Proposed Scheme and Comparison of Results sections describe an LSTM autoencoder that detects anomalous user behavior and achieves relatively good performance on insider-threat data. This supports deep learning's ability to identify patterns in that setting, but the paper does not establish the broader claim about intricate patterns across complex datasets generally. (LLM, conf=0.82)
- Not checked — no full text of the sourcecites [55] source
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“The authors of [62], [63] show that the effectiveness of these methods hinges on the quality of the baseline behavior models.”
What the sources say
- Not checked — no full text of the sourcecites [62] source
- Supportedcites [63] source
The Detection Mechanisms section (page unavailable) says anomaly-based methods define users’ normal activity patterns and flag deviations. The Limitations section (page unavailable) reports that insufficient features caused normal activities to be flagged as outliers, producing many false alarms. Together, these support the paraphrase that the effectiveness of behavior-based detection depends on how well normal behavior is modeled. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
16 more claims · 1 with their cited sources checked
Create a free account to see all 99 claims and to verify the 18 references still open.
- It may state two separate claims as one. 36
- Probably not a substantial claim. 64
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Therefore, despite the advances already achieved in LLM security, a significant gap remains to be filled: current models are susceptible to attacks, particularly in cases where they fail to correctly interpret the semantics of interactions [10].”
From Evaluation of NeMo Guardrails as a Firewall for User–LLM Interaction (2026) · cites this paper as [10]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“However, overreliance on domain-specific features limits generalizability, a gap highlighted in reviews of ML methods [19,34].”
From A systematic review on insider threat detection using natural language processing (2025) · cites this paper as [34]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Studies rarely justify dataset selection beyond availability [34], and the lack of real-time, multi-source datasets (e.g., integrating social media or chat logs [27]) restricts model robustness.”
From A systematic review on insider threat detection using natural language processing (2025) · cites this paper as [34]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“Digital Object Identifier 10.1109/ACCESS.2017.DOI”
§ I. INTRODUCTION
“Citation information: DOI 10.1109/ACCESS.2024.3369906”
Cited works that were retracted
Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.
A note, not an issue: for example, a cited work that was retracted only after this paper was published, which its authors could not have known.
In the paper
“The deployment of insider threat detection systems in real security environments demands a nuanced approach that considers not only technical aspects but also navigates the intricacies of organizational dynamics and regulatory landscapes [34], [39].”
Cited work 10.1109/mysurucon55714.2022.9972599 · retracted 2024-03-22 · notice 10.1109/mysurucon55714.2022.10478498 · this paper published 2024
Citations
Cites
1 retracted
Singh, R., & Deorari, R. (2022) · IEEE Mysore Sub Section International Conference (MysuruCon) · doi:10.1109/mysurucon55714.2022.9972599 · Cited by 3 · cited 2×
“The deployment of insider threat detection systems in real security environments demands a nuanced approach that considers not only technical aspects but also navigates the intricacies of organizational dynamics and regulatory landscapes [34], [39].”
BackgroundMethodRetractedAl-Shehari, T., & Alsowail, R. A. (2021) · Entropy · doi:10.3390/e23101258 · Cited by 193
Analyzing Data Granularity Levels for Insider Threat Detection Using Machine Learning
Le, D. C., Zincir-Heywood, N., & Heywood, M. I. (2020) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2020.2967721 · Cited by 191 · cited 1×
“Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks, designed to handle sequential data, excel in capturing temporal dependencies in user behavior, making them well-suited for insider threat detection [63], [77].”
BackgroundAnomaly Detection for Insider Threats Using Unsupervised Ensembles
Le, D. C., & Zincir-Heywood, N. (2021) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2021.3071928 · Cited by 147 · cited 2×
“…demonstrating efficacy in detecting subtle deviations indicative of insider threats [48], [50], [51], [76], Several studies [45], [49], [55], [57], [62] have demonstrated that deep learning, characterized by the use of neural networks with multiple layers, has remarkable capabilities…”
BackgroundMethodBehavioral Based Insider Threat Detection Using Deep Learning
Nasir, R., Afzal, M., Latif, R., & Iqbal, W. (2021) · IEEE Access · doi:10.1109/access.2021.3118297 · Cited by 146 · cited 2×
“In this context, the effectiveness of the Variational Autoencoder(VAE) is demonstrated in [49] in identifying internal threats with high accuracy.”
BackgroundModeling User Search Behavior for Masquerade Detection
Salem, M. B., & Stolfo, S. J. (2011) · Lecture Notes in Computer Science · doi:10.1007/978-3-642-23644-0_10 · Cited by 129 · cited 1×
“Dataset, introduced by Salem and Stolfo [88], comprises host-based events from 34 regular users and includes masquerade sessions conducted by 14 individuals.”
BackgroundMalicious Insider Attack Detection in IoTs Using Data Analytics
Khan, A. Y., Latif, R., Latif, S., Tahir, S., Batool, G., & Saba, T. (2020) · IEEE Access · doi:10.1109/access.2019.2959047 · Cited by 116 · cited 1×
“Additionally, autoencoders, a form of unsupervised deep learning, have shown promise in reconstructing normal patterns and identifying anomalies [78].”
BackgroundInsider Threat Detection Using Machine Learning Approach
Bin Sarhan, B., & Altwaijry, N. (2022) · Applied Sciences · doi:10.3390/app13010259 · Cited by 111 · cited 3×
“For privacy-sensitive scenarios, federated learning facilitates collaborative learning across decentralized datasets without compromising individual data privacy [4], [8].”
BackgroundInternet of Things Applications: Opportunities and Threats
Rahmani, A. M., Bayramov, S., & Kiani Kalejahi, B. (2022) · Wireless Personal Communications · doi:10.1007/s11277-021-08907-0 · Cited by 110 · cited 1×
“Employees now access organizational resources from various locations and devices, demanding adaptability in detection strategies [9].”
BackgroundA Review of Insider Threat Detection Approaches With IoT Perspective
Kim, A., Oh, J., Ryu, J., & Lee, K. (2020) · IEEE Access · doi:10.1109/access.2020.2990195 · Cited by 93 · cited 1×
“In an enterprise environment, security data often resides in disparate systems such as network logs, endpoint detection and response (EDR) solutions, user behavior analytics (UBA) platforms, and identity and access management (IAM) systems [91].”
Background
Cited by
Kheddar, H. (2025) · Information Fusion · doi:10.1016/j.inffus.2025.103347 · Cited by 161
Generative AI in cybersecurity: A comprehensive review of LLM applications and vulnerabilities
Ferrag, M. A., Alwahedi, F., Battah, A., Cherif, B., Mechri, A., Tihanyi, N., Bisztray, T., & Debbah, M. (2025) · Internet of Things and Cyber-Physical Systems · doi:10.1016/j.iotcps.2025.01.001 · Cited by 148
Evolving techniques in cyber threat hunting: A systematic review
Mahboubi, A., Luong, K., Aboutorab, H., Bui, H. T., Jarrad, G., Bahutair, M., Camtepe, S., Pogrebna, G., Ahmed, E., Barry, B., et al. (2024) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2024.104004 · Cited by 94
Anomaly-based threat detection in smart health using machine learning
Tabassum, M., Mahmood, S., Bukhari, A. A., Alshemaimri, B. K., Daud, A. Y., & Khalique, F. (2024) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-024-02760-4 · Cited by 112
A Comprehensive Review of AI’s Current Impact and Future Prospects in Cybersecurity
Siam, A. A., Alazab, M., Awajan, A. W., & Faruqui, N. (2025) · IEEE Access · doi:10.1109/access.2025.3528114 · Cited by 88
Metadata
- Authors
- Alzaabi, F. R., & Mehmood, A.
- Affiliations
- Abu Dhabi University (AE)
- ORCID
- Mehmood, Abid
- Venue
- IEEE Access
- Published
- 2024
- Volume
- 12
- Pages
- 30907-30927
- ISSN
- 2169-3536 (ISSN-L)
- OpenAlex
- W4392151675
- Semantic Scholar
- 98b3d89075ca3345cb2cc550a9c34ab9cf9add41
- Cited by
- 204
- References
- 94
- Citations by year
- 2026: 83 · 2025: 95 · 2024: 26
- FWCI
- 76.36
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- IEEE Access (publishedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Abu Dhabi University’s Office of Research and Grant Programs
- Abu Dhabi University’s office of sponsored programs in the United Emirates funded this endeavor (19300752)
- Topics
- Physical Sciences › Computer Science › Information Systems › Information and Cyber Security
- Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Keywords
- insider threat detection, malicious insiders, privileged access, deep learning, natural language processing, large language models, time series analysis, temporal patterns, human behavior analysis, cybersecurity, information security
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: A Review of Recent Advances, Challenges, and Opportunities in Malicious Insider Threat Detection Using Machine Learning Methods. https://app.cochat.ai/papers/10.1109/access.2024.3369906
- The paper
- Alzaabi, F. R., & Mehmood, A. (2024). A Review of Recent Advances, Challenges, and Opportunities in Malicious Insider Threat Detection Using Machine Learning Methods. IEEE Access. https://doi.org/10.1109/access.2024.3369906
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.