Evaluating model drift in machine learning algorithms
The DOI matches its Crossref record; 17 of the 17 references it cites were found in a registry or an index.
Abstract
Machine learning is rapidly emerging as a valuable technology thanks to its ability to learn patterns from large data sets and solve problems that are impossible to model using conventional programming logic. As machine learning techniques become more mainstream, they are being applied to a wider range of application domains. These algorithms are now trusted to make critical decisions in secure and adversarial environments such as healthcare, fraud detection, and network security, in which mistakes can be incredibly costly. They are also a critical component to most modern autonomous systems. However, the data driven approach utilized by these machine learning methods can prove to be a weakness if the data on which the models rely are corrupted by either nefarious or accidental means. Models that utilize on-line learning or periodic retraining to learn new patterns and account for data distribution changes are particularly susceptible to corruption through model drift. In modeling this type of scenario, specially crafted data points are added to the training set over time to adversely influence the system, inducing model drift which leads to incorrect classifications. Our work is focused on exploring the resistance of various machine learning algorithms to such an approach. In this paper we present an experimental framework designed to measure the susceptibility of anomaly detection algorithms to model drift. We also exhibit our preliminary results using various machine learning algorithms commonly found in intrusion detection research.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
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?
In the citing paper
“Deployed models must be monitored and potentially corrected for temporal instability (that is, statistical drift 95 ), which can be difficult with encrypted data or algorithms.”
From Secure, privacy-preserving and federated machine learning in medical imaging (2020) · cites this paper as [95]
Citations
Cites
A tutorial on hidden Markov models and selected applications in speech recognition
Rabiner, L. R. (1989) · Proceedings of the IEEE · doi:10.1109/5.18626 · Cited by 22,743 · cited 1×
“This algorithm was chosen again due to its relative simplicity as well as its common appearance in ML-IDS research [9][15][16].”
MethodEstimating the Support of a High-Dimensional Distribution
Schölkopf, B., Platt, J. C., Shawe-Taylor, J., Smola, A. J., & Williamson, R. C. (2001) · Neural Computation · doi:10.1162/089976601750264965 · Cited by 6,168
Poisoning Attacks against Support Vector Machines
Biggio, B., Nelson, B., & Laskov, P. (2012) · arXiv · doi:10.48550/arxiv.1206.6389 · Cited by 1,331
Intrusion detection by machine learning: A review
Tsai, C.-F., Hsu, Y.-F., Lin, C.-Y., & Lin, W.-Y. (2009) · Expert Systems with Applications · doi:10.1016/j.eswa.2009.05.029 · Cited by 969
The 1999 DARPA off-line intrusion detection evaluation
Lippmann, R., Haines, J. W., Fried, D. J., Korba, J., & Das, K. (2000) · Computer Networks · doi:10.1016/s1389-1286(00)00139-0 · Cited by 876
The security of machine learning
Barreno, M., Nelson, B., Joseph, A. D., & Tygar, J. D. (2010) · Machine Learning · doi:10.1007/s10994-010-5188-5 · Cited by 817
An Analysis of the 1999 DARPA/Lincoln Laboratory Evaluation Data for Network Anomaly Detection
Mahoney, M. V., & Chan, P. K. (2003) · Lecture Notes in Computer Science · doi:10.1007/978-3-540-45248-5_13 · Cited by 535
Exploiting machine learning to subvert your spam filter
Nelson, B. A., Barreno, M., Chi, F. J., Joseph, A. D., Rubinstein, B. I. P., Saini, U., Sutton, C. A., Tygar, J. D., & Xia, K. (2008) · Edinburgh Research Explorer (University of Edinburgh) · Cited by 481
Systematic Poisoning Attacks on and Defenses for Machine Learning in Healthcare
Mozaffari-Kermani, M., Sur-Kolay, S., Raghunathan, A., & Jha, N. K. (2015) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2014.2344095 · Cited by 383
McPAD: A multiple classifier system for accurate payload-based anomaly detection
Perdisci, R., Ariu, D., Fogla, P., Giacinto, G., & Lee, W. (2009) · Computer Networks · doi:10.1016/j.comnet.2008.11.011 · Cited by 269
Research works printed but not linked: 1
- biggio, 2013, Evasion atacks against machine learning at test time, Machine Learning and Knowledge Discovery in Databases, 387
Cited by
Secure, privacy-preserving and federated machine learning in medical imaging
Kaissis, G. A., Makowski, M. R., Rückert, D., & Braren, R. F. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-020-0186-1 · Cited by 1,676
“(5) Deployed models must be monitored and potentially corrected for temporal instability (that is, statistical drift 95 ), which can be difficult with encrypted data or algorithms.”
BackgroundHadi, M. U., tashi, Q. A., Qureshi, R., Shah, A., muneer, A., Irfan, M., Zafar, A., Shaikh, M. B., Akhtar, N., Wu, J., et al. (2023) · doi:10.36227/techrxiv.23589741.v4 · Cited by 103
Introduction to Artificial Intelligence and Machine Learning for Pathology
Harrison, J. H., Gilbertson, J. R., Hanna, M. G., Olson, N. H., Seheult, J. N., Sorace, J. M., & Stram, M. N. (2021) · Archives of Pathology & Laboratory Medicine · doi:10.5858/arpa.2020-0541-cp · Cited by 82
Barriers Related to AI Implementation in Supply Chain Management
Shrivastav, M. (2022) · Journal of Global Information Management · doi:10.4018/jgim.296725 · Cited by 64
“It is of paramount importance in AI to continuously monitor for drift which in turn require dedicated commitment and resources (Nelson et al., 2015).”
BackgroundHadi, M. U., tashi, Q. A., Qureshi, R., Shah, A., muneer, A., Irfan, M., Zafar, A., Shaikh, M. B., Akhtar, N., Wu, J., et al. (2023) · doi:10.36227/techrxiv.23589741.v3 · Cited by 53
Djarum, D. H., Ahmad, Z. A., & Zhang, J. (2023) · Applied Soft Computing · doi:10.1016/j.asoc.2023.110466 · Cited by 29
Ji, X., Salmon, R., Mulliqi, N., Khan, U., Wang, Y., Blilie, A., Olsson, H., Pedersen, B. G., Sørensen, K. D., Ulhøi, B. P., et al. (2025) · Modern Pathology · doi:10.1016/j.modpat.2025.100715 · Cited by 35
An, G. C., & Cockrell, C. (2023) · Frontiers in Systems Biology · doi:10.3389/fsysb.2023.1188009 · Cited by 18
“In these cases, the discrepancy between the training set and the eventual application in the real world leads to the condition termed data drift (Nelson, 2015; Baier et al., 2019; Ackerman et al, 2020).”
Development and Analysis of Multiscale Models for Tuberculosis: From Molecules to Populations
Nanda, P., Budak, M., Michael, C. T., Krupinsky, K. C., & Kirschner, D. E. (2024) · Modeling and simulation in science, engineering & technology · doi:10.1007/978-3-031-56794-0_2 · Cited by 14
Distributed Learning in Healthcare
Tuladhar, A., Rajashekar, D., & Forkert, N. D. (2022) · Integrated science · doi:10.1007/978-3-031-11199-0_10 · Cited by 11
The role of machine learning in autism spectrum disorder assessment and management
Reilly, A., Walsh, N., O’Reilly, D. P., Smyth, M., Gorman, K. M., Ostadabbas, S., & Power, C. (2025) · Pediatric Research · doi:10.1038/s41390-025-04566-0 · Cited by 17
“Moreover, diagnostic criteria continue to evolve 141 and developmental variability 183 and data drift over time 184 can degrade model performance unless systems are regularly updated and recalibrated.”
Jones, W. S., & Farrow, D. J. (2025) · Scientific Reports · doi:10.1038/s41598-025-94427-x · Cited by 9
“ML models may also be vulnerable to adversarial attacks, where intentional perturbations are introduced into the input data to manipulate model outputs 15,16 .”
Endogenous Macrodynamics in Algorithmic Recourse
Altmeyer, P., Angela, G., Buszydlik, A., Dobiczek, K., van Deursen, A., & Liem, C. C. S. (2023) · IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml54575.2023.00036 · Cited by 18
“review how resistant different machine learning models are to model drift [20].”
BackgroundCockrell, C., Schobel, S., Lisboa, F., Vodovotz, Y., & An, G. C. (2022) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2022.11.22.517524 · Cited by 11
“However, despite its promise, there are known fundamental and intrinsic limitations to ML, primarily related to challenges in generalizing to new use-contexts/cases and an inevitable performance degradation over time due to data drift [3-6].”
BackgroundThe Integration and Impact of Artificial Intelligence in Otolaryngology—Head and Neck Surgery
Crowson, M. G., & Nwosu, O. I. (2024) · Otolaryngologic Clinics of North America · doi:10.1016/j.otc.2024.04.001 · Cited by 9
Jones, A. C., Bandstra, M. S., Faaland, S., Lai, Y. S., Abgrall, N., Suchyta, S., & Cooper, R. J. (2025) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2025.3617856 · Cited by 5
“…in machine learning and artificial intelligence has shown that even for sophisticated models there is an inevitable drift over time away from optimal performance due to the widening gap between the data used to train the model and the current data on which it is being evaluated [47], [48], [49].”
Cloud-Native AI and Generative AI on AWS: A Systematic Review and a Proposed Unified Model
Shah, J., Gade, S. R., Ponna, D., & Patel, K. A. (2026) · International Conference on Mobile Computing and Sustainable Informatics (ICMCSI) · doi:10.1109/icmcsi67283.2026.11412536 · Cited by 4
Identifying runtime libraries in statically linked linux binaries
Carrillo-Mondéjar, J., & Rodríguez, R. J. (2024) · Future Generation Computer Systems · doi:10.1016/j.future.2024.107602 · Cited by 3
“Therefore, the e ff ectiveness of the machine learning model used in MANTILLA may degrade if the training data becomes outdated (Nelson et al., 2015).”
The Impact of Environment Configurations on the Stability of AI-Enabled Systems
Rahman, M., Khatoonabadi, S., Abdellatif, A., Sammaneh, H., & Shihab, E. (2025) · International Conference on Evaluation and Assessment in Software Engineering (EASE) · doi:10.1145/3756681.3756953 · Cited by 4
“Because AI-enabled systems are trained on existing data, any changes in the data cause model performance degradation over time [52].”
New Data Novelty Check and Distributed Learning for IoT Data Anomaly Detection
Jelassi, O., Negi, N., & Chaouchi, H. (2023) · Lecture notes in networks and systems · doi:10.1007/978-981-99-3236-8_9 · Cited by 2
Kutler, R. B., He, L., Green, R. W., & Rameau, A. (2025) · Current Opinion in Otolaryngology & Head & Neck Surgery · doi:10.1097/moo.0000000000001041 · Cited by 2
Test & Evaluation Best Practices for Machine Learning-Enabled SystemsPreprint
Chandrasekaran, J., Cody, T., McCarthy, N., Lanus, E., & Freeman, L. E. B. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4862589 · Cited by 14
Garcia-Rubio, A. M., Heck, L., Schweitzer, K. M., Bateman, R. M., Alaeddini, A., & Castillo-Villar, K. K. (2026) · IEEE Access · doi:10.1109/access.2026.3652106 · Cited by 2
“When these changes occur, the most straightforward way of alleviating it is often to simply update or retrain the object detection model [11], [13], [14].”
Measuring data drift with the unstable population indicator
Haas, M., & Sibbald, L. (2024) · Data Science · doi:10.3233/ds-240059 · Cited by 1
A Scalable Data Augmentation and Training Pipeline for Logo Detection
Guo, H., Swaminathan, V., & Mitra, S. (2019) · IEEE International Symposium on Multimedia (ISM) · doi:10.1109/ism46123.2019.00016 · Cited by 1
“Illustration of logo detection challenges SLST, as a completely automatic system, inevitably has the model drift problem [23] - the errors in model prediction can be propagated through the iterations.”
BackgroundDeriving and evaluating a fault model for testing data science applications
Jilani, A. A. A., Sherin, S., Ijaz, S., Iqbal, M. Z., & Khan, M. U. (2022) · Journal of Software Evolution and Process · doi:10.1002/smr.2449
“67 There are three main types of concept drift, that eventually lead to model decay: sudden drift, gradual drift, and incremental drift.”
BackgroundUncertainty Analysis in Machine Learning
Sisode, V., & Haribhakta, Y. V. (2023) · International Conference on Computing, Communication and Networking Technologies (ICCCNT) · doi:10.1109/icccnt56998.2023.10306377
“[4]In this research paper they proposed Confidence Difference (DoC).”
MethodOrchestrating Communication and Computation Efficiency in Edge-Based Knowledge Distillation
Lin, G., Feng, W., Luo, S., & Qian, B. (2024) · International Conference on Artificial Intelligence of Things and Systems (AIoTSys) · doi:10.1109/aiotsys63104.2024.10780538
Interpretable Drift Correction with Adaptive Transformation Selection
Shakhovska, K., & Pukach, P. Y. (2025) · Mathematical Modeling and Computing · doi:10.23939/mmc2025.04.1065
Gudibanda, A., Gupta, L. K., Mishra, R., Wadhera, T., & Sharma, N. (2025) · IEEE International Conference on Recent Trends in Computing and Smart Mobility (RCSM) · doi:10.1109/rcsm67767.2025.11507979
“This phenomenon is known as model drift [21] , which necessitates continuous and feasible updates to the model.”
Falowo, O. I., Abdo, J. B., & Abdo, J. B. (2026) · IEEE Middle East and North Africa Communications Conference (MENACOMM) · doi:10.1109/menacomm69507.2026.11532619
“The degradation of machine learning model performance over time due to changes in the underlying data distribution or concept drift [31].”
Metadata
- Authors
- Nelson, K., Corbin, G., Anania, M., Kovacs, M., Tobias, J., & Blowers, M.
- Affiliations
- United States Air Force Research Laboratory (US)
- Venue
- 2015 IEEE Symposium on Computational Intelligence for Security and Defense Applications (CISDA)
- Publisher
- IEEE
- Published
- 2015-05
- Pages
- 1-8
- OpenAlex
- W1487801014
- Cited by
- 31
- References
- 17
- Citations by year
- 2026: 3 · 2025: 8 · 2024: 6 · 2023: 7 · 2022: 4 · 2021: 1 · 2020: 1 · 2019: 1
- FWCI
- 0.00
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE Symposium on Computational Intelligence for Security and Defense Applications (CISDA) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Data Stream Mining Techniques
- Keywords
- model drift, anomaly detection, intrusion detection, machine learning robustness, data poisoning, online learning, distribution shift, classification accuracy
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.
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.