Enhancing Insider Threat Detection in Imbalanced Cybersecurity Settings Using the Density-Based Local Outlier Factor Algorithm
Open accessThe DOI matches its Crossref record; 33 of the 35 references it cites were found in a registry or an index; 6 of the 31 sources its claims cite were checked for support.
Abstract
This research introduces an approach using the Density-Based Local Outlier Factor algorithm to address class imbalance in the CERT r4.2 insider threat dataset.
Overview
CoChat Summary
TL;DR
The paper proposes a DBLOF-based anomaly detection approach for insider threat detection on the highly imbalanced CERT r4.2 dataset. With contamination set to 0.02, it reports 98% detection rate and 98.9% F1 score, outperforming the supervised baselines on threat detection metrics.
Key findings
- On the CERT r4.2 dataset, the authors report that the best DBLOF configuration at contamination rate 0.02 achieved 0.98 accuracy, 0.98 detection rate, and 0.99 F-score.
- The baseline supervised models performed much worse on detection and F1: XGB 0.159 / 0.27, Random Forest 0.246 / 0.36, Decision Tree 0.637 / 0.66, and KNN 0.217 / 0.36 (detection rate / F1 score).
- The paper reports that the dataset is highly skewed; one description says the CERT r4.2 set has 1,000 users with 7% malicious, while another states the training set contains 1,846,773 normal and 277 malicious instances and the test set 461,694 normal and 69 malicious instances.
- The authors conclude that accuracy alone is misleading in this imbalanced setting and that detection rate and F-score are more informative.
Methods
- The study uses the CERT r4.2 insider threat dataset and applies preprocessing including missing-value handling, feature extraction, feature scaling, one-hot encoding, and train/test partitioning.
- The proposed method is DBLOF (Density-Based Local Outlier Factor), applied as an anomaly-based / one-class approach to identify insider threats as local outliers.
- Performance is compared against supervised baselines: XGBoost, Random Forest, Decision Tree, and K-Nearest Neighbors, with DBLOF evaluated under multiple contamination ratios (0.00, 0.02, 0.04, 0.06, 0.08, 0.1).
Limitations
- Stated in the paper: the model is highly sensitive to the contamination rate, has generalizability and overfitting concerns, and may lack straightforward interpretability and scalability.
- Stated in the paper: the authors say the model has been validated on a specific dataset (CERT r4.2) and that its ability to generalize to other data is not yet verified.
- Evident from the text: some reported dataset descriptions and class counts appear internally inconsistent across sections, which complicates interpretation of the experimental setup.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
4 Valid · 4 Not valid · 22 Not checked · 90 Own results and interpretations
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“This method achieved an F-score for the employed algorithms as (LR: 67%, DT: 99%, RF: 99%, NB: 85%, KNN: 98%).”
What the sources say
- Could not verifycites [35] sourceExperimental — accuracy not yet measured
Section 4.3 discusses F-measure results under SMOTE but gives no LR F-score in the passages available from our copy; it reports an LR AUC-ROC value of 0.79, which is a different metric. Searches for “67%,” “0.67,” and LR F-measure did not locate the claimed 67% figure. (LLM, conf=0.72)
plain- 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
Every source the paper cites for this claim was checked and supports it.
In the paper
“Utilizing the RF classification algorithm and data sampling method, an accuracy score of 98% is obtained.”
What the sources say
- Supportedcites [34] Noever, 2019, Classifier suites for insider threat detection, arXiv:1901.10948 source
The Results report that random-forest implementations scored over 98% accuracy on the over-sampled dataset, supporting the cited 98% figure. The figure applies to over-sampled data; the paper reports 94% for under-sampled data. (LLM, conf=0.90)
plain- 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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This method achieved an F-score for the employed algorithms as (LR: 67%, DT: 99%, RF: 99%, NB: 85%, KNN: 98%).”
What the sources say
- Partly supportedcites [35] sourceExperimental — accuracy not yet measured
Section 4.3 (SMOTE) reports that KNN was among the algorithms whose precision, recall and F-measure scores were between 98% and 99%. This supports a high KNN F-score, but the passage does not establish that its F-score specifically was 98%. (LLM, conf=0.80)
plain- 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.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“This method achieved an F-score for the employed algorithms as (LR: 67%, DT: 99%, RF: 99%, NB: 85%, KNN: 98%).”
What the sources say
- Could not verifycites [35] sourceExperimental — accuracy not yet measured
Section 4.3 discusses F-measure results with SMOTE but does not give an NB F-score in the passages available from our copy. Section 5 reports an NB AUC-ROC value of 0.84, which is a different metric. Searches for “85%” and “Table 11 NB precision recall F-measure 0.85” did not locate the claimed 85% F-score. (LLM, conf=0.68)
plain- 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
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Through rigorous testing and validation processes, we have achieved outstanding results, with an of F-score 98%.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
59 more claims
Create a free account to see all 120 claims and to verify the 4 references still open.
- Probably not a substantial claim. 31
- The quote only partly states this claim. 4
- It may state two separate claims as one. 28
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 major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ V. Discussion
“Regarding the accuracy metric, we found that the baseline algorithms (XGB, RF, DT, and KNN) all achieved an accuracy score of 99%, whereas the suggested DBLOF detection algorithm approach only achieved an accuracy score of 98%.”
§ B) Anomaly-based Insider Threat Detection
“The highest detection results are achieved when the contamination ratio is set to 0.02 with an accuracy and detection rate of 99% and an F-score of 99% due to the contamination ratio being close to the minority class or outliers in the dataset.”
Citations
Cites
SMOTE synthetic minority over-sampling technique
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002) · Journal of Artificial Intelligence Research · doi:10.1613/jair.953 · Cited by 32,865 · cited 2×
“References should be numbered in order of appearance and indicated by a numeral or numerals in square brackets—e.g., [1] or [2], [3], or [4], [5], [6].”
BackgroundMethodLOF: identifying density-based local outliers
Breunig, M. M., Kriegel, H.-P., Ng, R. T., & Sander, J. (2000) · ACM SIGMOD Record · doi:10.1145/335191.335388 · Cited by 5,207
Learning from Imbalanced Data Sets
Fernández, A., García, S., Galar, M., Prati, R. C., Krawczyk, B., & Herrera, F. (2018) · doi:10.1007/978-3-319-98074-4 · Cited by 1,349 · cited 1×
“When the minority class is highly disorganized and made up largely of noisy examples or small disjoints, anomaly detection strategies excel [17].”
BackgroundA Comparative Evaluation of Unsupervised Anomaly Detection Algorithms for Multivariate Data
Goldstein, M., & Uchida, S. (2016) · PLOS ONE · doi:10.1371/journal.pone.0152173 · Cited by 1,032 · cited 1×
“≈ 1 Similar density as neighbors > 1 Lower density than neighbors ( Outlier ) 1 Higher density than neighburs ( Inlier ) (7) The most well-known density-based local-outlier detection algorithm is the DBLOF [28], which also introduced the idea of local outliers [29].”
BackgroundAn experimental comparison of performance measures for classification
Ferri, C., Hernández-Orallo, J., & Modroiu, R. (2009) · Pattern Recognition Letters · doi:10.1016/j.patrec.2008.08.010 · Cited by 889 · cited 1×
“The F-score is extensively utilized not just because of its expressive capacity but also because it is a notable metric for imbalanced classification [32].”
BackgroundHariri, S., Kind, M. C., & Brunner, R. J. (2021) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2019.2947676 · Cited by 501
Fernández, A., López, V., Galar, M., del Jesus, M. J., & Herrera, F. (2013) · Knowledge-Based Systems · doi:10.1016/j.knosys.2013.01.018 · Cited by 388 · cited 2×
“References should be numbered in order of appearance and indicated by a numeral or numerals in square brackets—e.g., [1] or [2], [3], or [4], [5], [6].”
BackgroundMethodBridging the Gap: A Pragmatic Approach to Generating Insider Threat Data
Glasser, J., & Lindauer, B. (2013) · 2013 IEEE Security and Privacy Workshops · doi:10.1109/spw.2013.37 · Cited by 344 · cited 1×
“The CERT dataset’s creation process is described in great length in [24].”
MethodA Survey of Recent Trends in One Class Classification
Khan, S. S., & Madden, M. G. (2010) · Lecture Notes in Computer Science · doi:10.1007/978-3-642-17080-5_21 · Cited by 330 · cited 1×
“When data labeling is expensive or one class happens considerably less often than others, this strategy may present certain concerns [26].”
BackgroundDeep learning for insider threat detection: Review, challenges and opportunities
Yuan, S., & Wu, X. (2021) · Computers & Security · doi:10.1016/j.cose.2021.102221 · Cited by 271 · cited 2×
“This scenario is inappropriate in the realm of threat detection since the minority class is essential to detection [2], [3].”
BackgroundHomoliak, I., Toffalini, F., Guarnizo, J., Elovici, Y., & Ochoa, M. (2020) · ACM Computing Surveys · doi:10.1145/3303771 · Cited by 255
Al-Shehari, T., & Alsowail, R. A. (2021) · Entropy · doi:10.3390/e23101258 · Cited by 193 · cited 1×
“Al-Shehari and Alsowail [35] proposed an insider data leakage model utilizing classification algorithms (LR, DT, RF, NB, KNN) combined with the SMOTE technique to address the imbalance issue of the dataset.”
MethodSelf-citationAnalyzing 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×
“The most widely used and easily available synthetic dataset for evaluating an insider threat detection system is the CERT dataset [18], which has been utilized in several insider threat detection studies such as [19], [20], [21], [22], and [23].”
MethodA New Take on Detecting Insider Threats: Exploring the Use of Hidden Markov Models
Rashid, T., Agrafiotis, I., & Nurse, J. R. C. (2016) · Proceedings of the 8th ACM CCS International Workshop on Managing Insider Security Threats · doi:10.1145/2995959.2995964 · Cited by 152 · cited 1×
“Rashid et al. [12] applied a hidden Markov model (HMM) in a recent study for insider anomalous detection.”
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 1×
“The most widely used and easily available synthetic dataset for evaluating an insider threat detection system is the CERT dataset [18], which has been utilized in several insider threat detection studies such as [19], [20], [21], [22], and [23].”
MethodAutomated Insider Threat Detection System Using User and Role-Based Profile Assessment
Legg, P. A., Buckley, O., Goldsmith, M., & Creese, S. (2017) · IEEE Systems Journal · doi:10.1109/jsyst.2015.2438442 · Cited by 145 · cited 1×
“A framework for insider threat problem formulation was proposed by Legg et al. [14].”
Behavioral Analysis of Insider Threat: A Survey and Bootstrapped Prediction in Imbalanced Data
Azaria, A., Richardson, A., Kraus, S., & Subrahmanian, V. S. (2014) · IEEE Transactions on Computational Social Systems · doi:10.1109/tcss.2014.2377811 · Cited by 133 · cited 2×
“This scenario is inappropriate in the realm of threat detection since the minority class is essential to detection [2], [3].”
BackgroundHow Data Scientists Use Computational Notebooks for Real-Time Collaboration
Wang, A. Y., Mittal, A., Brooks, C., & Oney, S. (2019) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3359141 · Cited by 131 · cited 1×
“Because of the many advantages it provides for the facilitation of the usage of code to handle a number of issues [31], businesses are increasingly embracing Google’s Colab as an ICT for data science.”
BackgroundDetecting insider threats in a real corporate database of computer usage activity
Senator, T. E., Goldberg, H. G., Memory, A., Young, W. T., Rees, B., Pierce, R., Huang, D., Reardon, M., Bader, D. A., Chow, E., et al. (2013) · Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/2487575.2488213 · Cited by 130 · cited 1×
“The approaches in [8] and [9] applied different anomaly detection techniques (e.g., hidden Markov model (HMM), Gaussian Mixture Model (GMM), etc.).”
MethodMulti-Domain Information Fusion for Insider Threat Detection
Eldardiry, H., Bart, E., Liu, J., Hanley, J., Price, B., & Brdiczka, O. (2013) · 2013 IEEE Security and Privacy Workshops · doi:10.1109/spw.2013.14 · Cited by 100 · cited 1×
“In [10], a hybrid anomaly method was proposed for detecting blend-in outliers and unusual change anomalies.”
MethodGavai, G., Sricharan, K., Gunning, D., et al. (2015) · J. Wirel. Mob. Networks Ubiquitous Comput. Dependable Appl. · doi:10.22667/jowua.2015.12.31.047 · Cited by 96
Detecting Insider Threat from Enterprise Social and Online Activity Data
Gavai, G., Sricharan, K., Gunning, D., Rolleston, R., Hanley, J., & Singhal, M. (2015) · Proceedings of the 7th ACM CCS International Workshop on Managing Insider Security Threats · doi:10.1145/2808783.2808784 · Cited by 69 · cited 1×
“An anomaly and quitter detection mechanism were proposed to compact insider threats by Gavai et al. [11] They employed various machine learning-based techniques on insiders’ activity log data throughout the organization.”
MethodEvaluating Insider Threat Detection Workflow Using Supervised and Unsupervised Learning
Le, D. C., & Zincir-Heywood, A. N. (2018) · 2018 IEEE Security and Privacy Workshops (SPW) · doi:10.1109/spw.2018.00043 · Cited by 66 · cited 1×
“In [13], a model for categorizing and presenting malicious and normal users’ activities within an organization was proposed.”
BackgroundImage-Based Feature Representation for Insider Threat Classification
Gayathri, R. G., Sajjanhar, A., & Xiang, Y. (2020) · Applied Sciences · doi:10.3390/app10144945 · Cited by 63
An Integrated Imbalanced Learning and Deep Neural Network Model for Insider Threat Detection
Al-Mhiqani, M. N., Ahmed, R., Zainal, Z., & Isnin, S. N. (2021) · International Journal of Advanced Computer Science and Applications · doi:10.14569/ijacsa.2021.0120166 · Cited by 63 · cited 1×
“The most widely used and easily available synthetic dataset for evaluating an insider threat detection system is the CERT dataset [18], which has been utilized in several insider threat detection studies such as [19], [20], [21], [22], and [23].”
MethodGraph Based Framework for Malicious Insider Threat Detection
Gamachchi, A., Sun, L., & Boztas, S. (2017) · Proceedings of the Annual Hawaii International Conference on System Sciences · doi:10.24251/hicss.2017.319 · Cited by 59
Exploring anomalous behaviour detection and classification for insider threat identification
Le, D. C., & Zincir‐Heywood, N. (2021) · International Journal of Network Management · doi:10.1002/nem.2109 · Cited by 56 · cited 1×
“The most widely used and easily available synthetic dataset for evaluating an insider threat detection system is the CERT dataset [18], which has been utilized in several insider threat detection studies such as [19], [20], [21], [22], and [23].”
MethodSupervised Learning for Insider Threat Detection Using Stream Mining
Parveen, P., Weger, Z. R., Thuraisingham, B., Hamlen, K., & Khan, L. (2011) · 2011 IEEE 23rd International Conference on Tools with Artificial Intelligence · doi:10.1109/ictai.2011.176 · Cited by 45 · cited 1×
“The One-Class Support Vector Machine (OCSVM) method was suggested by Parveen et al. [15] for the purpose of classifying data into normal and abnormal categories, specifically for the detection of harmful insider threats.”
MethodEmpirical Detection Techniques of Insider Threat Incidents
Alsowail, R. A., & Al-Shehari, T. (2020) · IEEE Access · doi:10.1109/access.2020.2989739 · Cited by 41 · cited 2×
“The recent survey in [6] summarized the insider threat detection models in the literature.”
BackgroundSelf-citationEVOLVING INSIDER THREAT DETECTION STREAM MINING PERSPECTIVE
PARVEEN, P., MCDANIEL, N., WEGER, Z., EVANS, J., THURAISINGHAM, B., HAMLEN, K., & KHAN, L. (2013) · International Journal on Artificial Intelligence Tools · doi:10.1142/s0218213013600130 · Cited by 26 · cited 1×
“The authors then expanded upon their previous research in a subsequent publication [16], whereby they used the ensemble technique to investigate the application of unsupervised Graph-Based Anomaly Detection (GBAD).”
MethodAl-Shehari, T., & Alsowail, R. A. (2023) · International Journal of Information Security · doi:10.1007/s10207-022-00651-1 · Cited by 25
Self-citationArtificial Intelligence and Cognitive Science
Coyle, L., & Freyne, J. (2010) · Lecture notes in computer science · doi:10.1007/978-3-642-17080-5 · Cited by 24
Insider Threat Detection in PRODIGAL
Goldberg, H., Young, W., Reardon, M., Phillips, B., & Senator, T. (2017) · Proceedings of the Annual Hawaii International Conference on System Sciences · doi:10.24251/hicss.2017.320 · Cited by 20 · cited 1×
“The approaches in [8] and [9] applied different anomaly detection techniques (e.g., hidden Markov model (HMM), Gaussian Mixture Model (GMM), etc.).”
MethodClassifier Suites for Insider Threat Detection
Noever, D. (2019) · arXiv · doi:10.48550/arxiv.1901.10948 · Cited by 16 · cited 1×
“In [34], a classification model for insider threat detection was proposed based on the RF.”
MethodTowards an Explainable Approach for Insider Threat Detection: Constraint Network Learning
Orizio, R., Vuppala, S., Basagiannis, S., & Provan, G. (2020) · 2020 International Conference on Intelligent Data Science Technologies and Applications (IDSTA) · doi:10.1109/idsta50958.2020.9264049 · Cited by 10
Research works printed but not linked: 3
- 2022, Greatest Threat
- 2020, Insider Threat Test Dataset
- Ertoz, 2004, Minds-minnesota intrusion detection system, Next Generation Data Mining
Cited by
Dhanushkodi, K., & Thejas, S. (2024) · IEEE Access · doi:10.1109/access.2024.3493957 · Cited by 130
Mladenovic, D., Antonijević, M., Jovanović, L., Šimić, V., Zivkovic, M. Z., Bačanin, N., Živković, T., & Perišić, J. (2024) · Scientific Reports · doi:10.1038/s41598-024-77240-w · Cited by 102
“This dataset consists of several scenarios of simulated intrusion events, and was utilized in numerous research papers including 55–58 .”
MethodApplications of Machine Learning in Cyber Security: A Review
Vourganas, I., & Michala, A. L. (2024) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp4040045 · Cited by 35
Nalini, M., Yamini, B., Ambhika, C., & Subramanian, R. S. (2024) · International Journal of Machine Learning and Cybernetics · doi:10.1007/s13042-024-02460-5 · Cited by 18
“Through ML algorithms, blockchain technology can effectively analyze vast amounts of data, enhancing security and providing valuable insights [3, 10].”
Prasad, P. S. S., Nayak, S. K., & Krishna, M. V. (2025) · International Research Journal of Multidisciplinary Technovation · doi:10.54392/irjmt2527 · Cited by 26
Metadata
- Authors
- Al-Shehari, T. A., Rosaci, D., Al-Razgan, M., Alfakih, T., Kadrie, M., Afzal, H., & Nawaz, R.
- Affiliations
- Department of Self-Development Skill, Common First Year Deanship, King Saud University, Riyadh, Saudi Arabia
- University of Reggio Calabria (IT)
- Department of Software Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia
- King Saud University (SA)
- National University of Sciences and Technology (PK)
- University of Staffordshire (GB)
- ORCID
- Venue
- IEEE Access
- Published
- 2024
- Volume
- 12
- Pages
- 34820-34834
- ISSN
- 2169-3536 (ISSN-L)
- OpenAlex
- W4392406063
- Cited by
- 51
- References
- 35
- Citations by year
- 2026: 27 · 2025: 19 · 2024: 5
- FWCI
- 8.52
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- IEEE Access (publishedVersion, cc-by-nc-nd)
- Staffordshire Online Repository (Staffordshire University) (submittedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- King Saud University, Riyadh, Saudi Arabia, through the Researchers Supporting Project (RSP2024R206)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
- Physical Sciences › Computer Science › Information Systems › Information and Cyber Security
- Physical Sciences › Computer Science › Artificial Intelligence › Anomaly Detection Techniques and Applications
- Keywords
- insider threat detection, class imbalance, anomaly detection, false negative reduction, cybersecurity
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).
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
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.