Pith. sign in

REVIEW 3 major objections 5 minor 69 references

An Unsupervised Anomaly Detection in Electricity Consumption Using Reinforcement Learning and Time Series Forest Based Framework

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read An RL-trained model picker, choosing the right detector at each time step, can beat every individual anomaly detector on real electricity-consumption data.

desk verdict The TSF-reward twist and the anomaly-type comparison are worth a look, but the central accuracy claim rests on training and scoring the selector on the same anomalous sequence, so the reported F1 is not evidence of generalization. read the letter →

arxiv 2501.00107 v1 pith:CUZPXVKS submitted 2024-12-30 cs.LG cs.AI

Jihan Ghanim , Mariette Awad This is my paper · ORCID
classification cs.LGcs.AI
keywords anomalydetectiontimeseriesreinforcementlearningmodelselectionforestDQNelectricityconsumptionunsupervised
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that an agent trained with reinforcement learning can decide, at every time step, which of six unsupervised anomaly detectors to trust for electricity-consumption data, using only a small fraction of ground-truth labels. The proposed selector dynamically picks among KNN, COPOD, ECOD, one-class SVM, isolation forest, and USAD, guided by six time-series-forest classifiers that estimate whether each detector's prediction is correct. On the real smart-meter dataset the paper reports a precision of 0.977 and an F1 score of 0.727, beating every individual detector in precision and F1; on a synthetic dataset it reaches an F1 of 0.989, second only to KNN's perfect 1.0. If this holds, operators can obtain adaptive anomaly detection without large labeled sets, and can keep high performance across global, local, and clustered anomaly types.

What carries the argument

The load-bearing object is the DQN policy combined with six TSF classifiers, one per candidate detector. Each TSF is trained on 20% of the anomalous sequence to classify whether its detector's predicted label is correct, using input features that include the window, the rescaled anomaly score, the empirical threshold, the predicted label, and two confidence scores: distance-to-threshold and prediction consensus. The DQN uses these TSF outputs as pseudo-labels for most reward computation, so the agent learns to select the detector most likely to be right without needing full ground truth. The asymmetric reward function rewards true positives more than true negatives and penalizes false negatives more than false positives.

What would settle it

Run the trained policy on a held-out anomalous sequence from meters not used in training, rewarding it only through the TSF classifiers, and compare its F1 with KNN's; a result at or below KNN's 0.702 on the real dataset would contradict the claim that the selector outperforms all candidate detectors.

Watch

Extended reading notes

Core claim

The central claim is that model selection for unsupervised anomaly detection can be learned as a sequential decision problem: a deep Q-network observes the current sliding window, all six detectors' anomaly scores, their empirical thresholds, predicted labels, and two confidence scores, then picks one detector; the reward compares the chosen detector's label with a TSF classifier's estimate of correctness on 80% of points and with ground truth on the remaining 20%. The authors report that this selector outscores all six candidate detectors on the real dataset and all but KNN on the synthetic dataset, and that it stays accurate across datasets separately injected with global, local, and clustered anomalies. The intended message is that a detector pool plus a modest label budget can replace the search for a single universally best algorithm.

Load-bearing premise

The reported outperformance assumes that the F1 scores, measured on the same anomalous dataset the DQN was trained on, carry over to new time series; the policy's rewards come from TSF predictions on 80% of that dataset and ground truth on the same 20% used to train the TSFs.

Editorial extensions

If this is right

  • A pool of imperfect detectors plus a learned switcher can outperform the best single detector in the pool on real electricity data.
  • Only about 20-30% of anomalous instances need labels to train the selector, relaxing the label bottleneck in unsupervised anomaly detection deployments.
  • Detector rankings change with anomaly type (global, local, clustered), so a fixed best model is not reliable across settings.
  • Reward design matters: the original asymmetric reward with decaying exploration beats both constant and adaptive reward alternatives.
  • On the synthetic benchmark, the RL selector exceeds a chain-of-thought prompted LLM anomaly detector (F1 0.989 vs 0.913).

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the paper's headline numbers are computed on the same anomalous sequence used to train the DQN, the practical advantage would need confirmation on a held-out sequence; the 20% label subset is also reused for TSF training and reward, so the effective label cost may be higher than stated.
  • The selector may be learning a meta-signal about detector reliability rather than the anomaly structure itself; if so, retraining on a new dataset's detector outputs could work even when the raw electricity patterns differ.
  • The TSF pseudo-labelers could be replaced by any calibrated classifier, making the scheme a general 'detector of detectors' approach applicable beyond electricity data.
  • A direct testable extension is to vary the 20% label fraction downward and measure the precision/F1 trade-off, which would map how much label savings the method truly buys.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes RLAD, a model-selection framework for time-series anomaly detection (AD) in electricity-consumption data. Six unsupervised AD models (KNN, COPOD, ECOD, OSVM, IForest, USAD) are trained on normal data, and six Time Series Forest (TSF) classifiers are trained on 20% of the anomalous test sequence to predict whether each AD model's prediction is correct. A DQN agent then chooses, at each time step, which AD model's label to output; its reward is computed from TSF predictions on 80% of the anomalous sequence and ground-truth labels on the same 20% used for TSF training. The paper reports F1 scores on a real LEAD1.0 subset and a synthetic dataset, claiming the proposed selector outperforms all candidate AD models on the real data and all but KNN on the synthetic data, and also outperforms a GPT-4-based detector on the synthetic data. Additional experiments study reward-function variations, exploration epsilons, and anomaly types (global, local, clustered).

Significance. If the reported results were valid, the paper would offer a useful recipe for dynamic AD model selection with limited labels, and its comparison against an LLM-based detector is topical. The paper also contains a reasonably broad survey of unsupervised AD categories and a systematic ablation of reward designs. However, the paper's central quantitative claim rests on a flawed evaluation protocol: the DQN policy is trained and scored on the same anomalous sequence, with TSF proxies that are themselves trained on ground-truth labels from that sequence. Because the TSFs attain F1 scores of 0.91-0.98 on the real data, the reward signal is nearly equivalent to giving the agent the test-set labels during training, while the six baselines receive no labels at all. The reported F1 advantage is therefore evidence of label leakage and in-sample optimization rather than of model-selection quality. No code or data is provided to reproduce the experiments, and the claimed generalization to unseen time series is unsupported.

major comments (3)
  1. [Secs. 3.4, 3.5, 4.4, 5.3] The evaluation trains and tests on the same anomalous dataset, so the central comparison is invalid. Section 3.4 trains six TSF classifiers on 20% of Danomaly using ground-truth labels. Section 3.5 defines the DQN reward as TSF predictions on 80% of Danomaly and direct ground truth on the same 20% used for TSF training, explicitly reusing those labels. Section 4.4 trains the DQN for 3,000,000 time steps on the real anomalous dataset, and Table 2 reports precision/F1 on that same dataset. Since the TSFs achieve F1 0.913-0.979 on this data (Table 4), the agent effectively receives near-perfect label information across the entire training sequence; the reported F1 of 0.727 on the real data reflects in-sample optimization, not model-selection quality. The six unsupervised baselines receive no labels, so the comparison conflates the selector's contribution with test-set labels. A held-out test sequence, with TSF training and DQN rewards confined to a disjunct training portion, is required to support any generalization claim.
  2. [Secs. 5.3 and 5.5, Table 7] The claimed superiority over unsupervised candidates is not robust even within the leaked evaluation. Table 7 shows that RLAD_Gtruth, which uses the full ground truth in its reward, achieves F1=0.793 on the real dataset while the proposed framework achieves F1=0.727. This indicates that the label-derived reward component is doing much of the work, and the proposed method does not even beat a fully label-rewarded variant. The paper should additionally compare against a semi-supervised baseline trained directly on the same 20% labels and evaluated on the remaining 80% (with the DQN likewise trained only on a training split), to demonstrate that the selector adds value beyond simply using the labels.
  3. [Sec. 4.3] The anomaly threshold for every candidate AD model is set to the 5% rank criterion, using the known anomaly percentage of each dataset. This transfers label-derived information (the global anomaly rate) into every baseline's decision rule, weakening the claim that the six candidate models are unsupervised. The paper reports no sensitivity analysis for this criterion, so it is unclear how much of the baseline performance, and of the proposed method's advantage, depends on this choice. The threshold should be treated as a hyperparameter and varied, or selected using only the normal training data.
minor comments (5)
  1. [Sec. 3.5] The reward specification text says 'rFN is set to -1.5, and rFN is set to -3'; the first value should be rFP = -1.5 and the second rFN = -3, as reflected in Equation (6).
  2. [Sec. 5.7.1] The text refers to 'the same decaying exploration epsilon of Figure 2', but Figure 2 is the LLM prompt template; the decaying-epsilon schedule is presumably the one described in Section 4.4. The figure cross-references should be corrected.
  3. [Sec. 5.8] The text says 'as shown in Tables 14' but the referenced table is Table 14; also, the table caption 'F1score of AD Model' is ungrammatical and the table header 'Global Local Clustered' should be marked as F1 scores.
  4. [Sec. 5.4] The GPT-4 comparison is made only on the synthetic dataset, where KNN already achieves F1=1; this comparison is not informative about the proposed method's value on the real dataset, and no statistical significance or variance across prompts is reported.
  5. [Abstract and Sec. 3] The abstract claims the approach works 'without explicitly depending on ground truth labels', but Section 3.4 requires 20% labels for TSF training and Section 3.5 uses labels in the reward. The wording should be revised to 'with a limited portion of ground truth labels'.

Circularity Check

1 steps flagged · score 6.0 of 10

The reported F1 on the real dataset is computed on the same anomalous sequence used to train the TSFs and DQN policy, so the claimed superiority over unsupervised baselines reflects in-sample optimization rather than a prediction on new data.

  1. fitted input called prediction [Sections 3.4-3.5 (TSF training and DQN reward), 4.4 (DQN training), 5.3 (reported performance)]
    "It should be mentioned that the same 20% of ground truth labels used during the training of the TSFs, are used here in the reward function. ... The agent was trained for 3,000,000 time steps in the real anomalous dataset case. ... results in Table 2 show that the proposed AD model selection, having a precision of 0.977 and an F1score of 0.727, outperformed all the candidate AD techniques of Table in terms of the precision and F1score."

    The DQN policy whose performance is reported is trained on the same Danomaly used for the reported F1. The reward function uses TSF predictions on 80% of Danomaly and the same 20% ground-truth labels used to train the TSFs on the remaining 20%. The TSFs themselves are trained on 20% of Danomaly and achieve F1 0.91-0.98 on the other 80%, so the reward signal is a high-fidelity proxy for the test labels. Thus the 'prediction' on the real dataset is an in-sample evaluation of a policy fitted to that exact sequence; the comparison with unsupervised baselines, which receive no Danomaly labels, is not a test of model-selection generalization but of fitting to test data.

full rationale

The framework is not definitionally circular: the TSF reward proxy is a legitimate design idea, and the DQN policy could in principle be evaluated on held-out data. The paper, however, never provides such a split. Section 3.4 trains the six TSF classifiers on 20% of Danomaly with ground truth labels, Section 3.5 explicitly reuses the same 20% labels in the DQN reward and uses TSF predictions on the remaining 80%, and Section 4.4 trains the DQN for 3,000,000 time steps on the real anomalous dataset whose F1 is then reported in Section 5.3. The candidate AD models, by contrast, are trained only on Dnormal and tested on Danomaly, so they receive no test labels. Because the TSFs attain F1 scores of 0.91-0.98 on this same data, the DQN's reward is nearly equivalent to seeing the test labels, making the reported advantage over the baselines a measure of in-sample fitting rather than a prediction on unseen time series. This is the central empirical claim of the paper, so the circularity score reflects that the headline result reduces to evaluating a fitted policy on its own training distribution. No load-bearing self-citation or uniqueness argument is involved; the issue is purely the self-referential evaluation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on hand-set reward values, a known anomaly rate used for thresholds, and an in-sample evaluation protocol where the learned policy is both trained and scored on the same anomalous sequence.

free parameters (5)
  • Reward values (rTN, rFN, rTP, rFP) = 0.5, -3, 1, -1.5
    Hand-chosen in Eq. (6) to prioritize TP and penalize FN; other values explored in Sec. 5.6 give lower F1.
  • Anomaly threshold criterion = 5%
    Set to the known anomaly percentage; used to binarize anomaly scores for all AD models (Sec. 4.3).
  • Sliding window size and step = size=6, step=1
    Chosen heuristically over sizes 6,12,24 and steps 1,2,3 (Sec. 4.2).
  • Exploration fraction = 0.7
    Set instead of default 0.1 to increase exploration in DQN (Sec. 4.4).
  • Ground truth split for TSF training = 20%
    The portion of anomalous ground truth labels used to train TSFs and in the reward function (Sec. 3.4, 3.5).
assumptions (4)
  • domain assumption The state features (anomaly scores, thresholds, labels, confidence scores) are sufficient for the RL agent to choose the best AD model at each time step.
    No evidence that these features capture all relevant information; this is assumed in the MDP design (Sec. 3.5).
  • domain assumption The TSF classifiers can accurately predict whether an AD model's prediction is correct, with accuracy high enough to shape the RL reward.
    TSF F1 scores are reported on an 80% holdout split (Tables 3,4), but no analysis connects TSF accuracy to final RL performance.
  • ad hoc to paper The RL policy trained on a specific anomalous dataset is a valid estimator of performance on that same dataset.
    The agent is trained for 600k/3M steps on D_anomaly and evaluated on the same D_anomaly; this is an in-sample evaluation, not a held-out test (Sec. 3.5, 5.3).
  • domain assumption The percentage of anomalies in the test data is known in advance and used to set detection thresholds.
    The 5% threshold criterion uses the known anomaly rate (Sec. 4.3), which may not be available in practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Unsupervised Anomaly Detection in Electricity Consumption Using Reinforcement Learning and Time Series Forest Based Framework." pith.science (2026). https://pith.science/paper/CUZPXVKS

@misc{pith2026250100107,
  author       = {Pith},
  title        = {Pith review of: An Unsupervised Anomaly Detection in Electricity Consumption Using Reinforcement Learning and Time Series Forest Based Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CUZPXVKS}},
  note         = {Machine review of arXiv:2501.00107}
}
read the original abstract

Anomaly detection (AD) plays a crucial role in time series applications, primarily because time series data is employed across real-world scenarios. Detecting anomalies poses significant challenges since anomalies take diverse forms making them hard to pinpoint accurately. Previous research has explored different AD models, making specific assumptions with varying sensitivity toward particular anomaly types. To address this issue, we propose a novel model selection for unsupervised AD using a combination of time series forest (TSF) and reinforcement learning (RL) approaches that dynamically chooses an AD technique. Our approach allows for effective AD without explicitly depending on ground truth labels that are often scarce and expensive to obtain. Results from the real-time series dataset demonstrate that the proposed model selection approach outperforms all other AD models in terms of the F1 score metric. For the synthetic dataset, our proposed model surpasses all other AD models except for KNN, with an impressive F1 score of 0.989. The proposed model selection framework also exceeded the performance of GPT-4 when prompted to act as an anomaly detector on the synthetic dataset. Exploring different reward functions revealed that the original reward function in our proposed AD model selection approach yielded the best overall scores. We evaluated the performance of the six AD models on an additional three datasets, having global, local, and clustered anomalies respectively, showing that each AD model exhibited distinct performance depending on the type of anomalies. This emphasizes the significance of our proposed AD model selection framework, maintaining high performance across all datasets, and showcasing superior performance across different anomaly types.

Figures

Figures reproduced from arXiv: 2501.00107 by the authors.

Figure 1
Figure 1. Illustration of the Proposed Framework where the elements of the confusion matrix are: True Positive (TP): represents the case where the anomalous instance is correctly predicted. False Positive (FP): stands for the case where the anomalous instance is predicted as a normal in￾stance. True Negative (TN): refers to the case where the normal instance is correctly predicted as a normal instance. False Negative (FN): re… view at source ↗
Figure 2
Figure 2. Template for the prompt strategy for the [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. The AdapInc reward is formulated in the equation below: RAdapInc =    rTN = 5×10−2 ×C 2 TN rFN = −3×10−2 ×C 2 FN rTP = 1×10−2 ×C 2 TP rFP = −1.5×10−2 ×C 2 FP (7) In both adaptive rewards, the counter C was ini￾tially set to 1, for every 100 time-steps the counter would be incremented by 1. When reaching the fi￾nal time-step (final sliding window of the anoma￾lous dataset), C would be reset to 1. Alternati… view at source ↗
Figures from the paper (2 more)
Figure 3
Figure 3. Figure 3: Adaptive Increasing Reward [PITH_FULL_IMAGE:figures/full_fig_p016_3.png]
Figure 4
Figure 4. Figure 4: Adaptive Decreasing Reward 5.8 Different Types of Anomalies In time series data, there exist different types of anomalies, consisting of global anomalies, local anomalies, and clustered anomalies. Global anomalies: point anomalies that seem to differ from the overall t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

69 extracted references · 64 canonical work pages

  1. [1]

    J. E. Zhang, D. Wu, and B. Boulet, Time Series Anomaly Detection via Reinforcement Learning-Based Model Selection, 2022 IEEE Canadian Conference on Electrical and Com- puter Engineering (CCECE), 2022, 193–199

    work page 2022

  2. [2]

    Alkuwari, S

    A.N. Alkuwari, S. Al-Kuwari, M. Qaraqe, Anomaly Detection in Smart Grids: A Survey from Cybersecurity Perspective, 2022 3rd In- ternational Conference on Smart Grid and Re- newable Energy (SGRE), 2022, 1-7

    work page 2022

  3. [3]

    Chatterjee, B.S

    A. Chatterjee, B.S. Ahmed, IoT Anomaly De- tection Methods and Applications: A Survey, Internet of Things, 19, 2022, 100568

    work page 2022

  4. [4]

    Khalil, F

    A.A. Khalil, F. E Ibrahim, M.Y . Abbass, N. Haggag, Y . Mahrous, A. Sedik, Z. Elsherbeeny, A.M. Khalaf, M. Rihan, W. El-Shafai, Effi- cient Anomaly Detection from Medical Signals and Images with Convolutional Neural Net- works for Internet of Medical Things (IoMT) Systems, International Journal for Numerical Methods in Biomedical Engineering, 38(1), 2022, e3530

    work page 2022

  5. [5]

    Hilal, S.A

    W. Hilal, S.A. Gadsden, J. Yawney, Financial Fraud: A Review of Anomaly Detection Tech- niques and Recent Advances, Expert Systems with Applications, 193, 2022, 116429

    work page 2022

  6. [6]

    M. U. Hassan, M. H. Rehmani, and J. Chen, Anomaly Detection in Blockchain Networks: A Comprehensive Survey, IEEE Communica- tions Surveys & Tutorials, 25(1), 2022, 289– 318

    work page 2022

  7. [7]

    Singh and K

    A. Singh and K. Chatterjee, Cloud Security Is- sues and Challenges: A Survey, Journal of Net- work and Computer Applications, 79, 2017, 88–115

    work page 2017

  8. [8]

    D. Jung, N. Ramanan, M. Amjadi, S. R. Karingula, J. Taylor, and C. N. Coelho Jr, Time Series Anomaly Detection with Label-Free Model Selection, arXiv preprint arXiv:2106.07473, 2021

    work page Pith review arXiv 2021

Show all 69 references
  1. [9]

    Barnett and T

    V . Barnett and T. Lewis, Outliers in Statistical Data, 3rd ed., Wiley, New York 1994

    1994

  2. [10]

    L. Ruff, J. R. Kauffmann, R. A. Vandermeulen, G. Montavon, W. Samek, M. Kloft, T. G. Di- etterich, and K.-R. M ¨uller, A Unifying Review of Deep and Shallow Anomaly Detection, Pro- ceedings of the IEEE, 109(5), 2021, 756–795

    2021

  3. [11]

    Gunduz and A

    M. Gunduz and A. M. A. Yahya, Analysis of Project Success Factors in Construction Indus- try, Technological and Economic Development of Economy, 24(1), 2018, 67–80

    2018

  4. [12]

    P. M. Tehrani, Cyber Resilience Strategy and Attribution in the Context of International Law, European Conference on Cyber Warfare and Security, 2019, 501–XVI

    2019

  5. [13]

    Ghanim, M

    J. Ghanim, M. Issa, and M. Awad, An Asym- metric Loss with Anomaly Detection LSTM Framework for Power Consumption Prediction, 2022 IEEE 21st Mediterranean Electrotechni- cal Conference (MELECON), 2022, 819–824

    2022

  6. [14]

    N. B. Aissa and M. Guerroumi, Semi- Supervised Statistical Approach for Network 20 Anomaly Detection, Procedia Computer Sci- ence, 83, 2016, 1090–1095

    2016

  7. [15]

    Akcay, A

    S. Akcay, A. Atapour-Abarghouei, and T. P. Breckon, Ganomaly: Semi-Supervised Anomaly Detection via Adversarial Training, Computer Vision–ACCV 2018: 14th Asian Conference on Computer Vision, Perth, Aus- tralia, December 2–6, 2018, Revised Selected Papers, Part III 14, 2019, 622–637

    2018

  8. [16]

    S. Han, X. Hu, H. Huang, M. Jiang, and Y . Zhao, Adbench: Anomaly Detection Bench- mark, Advances in Neural Information Pro- cessing Systems, 35, 2022, 32142–32159

    2022

  9. [17]

    Chandola, A

    V . Chandola, A. Banerjee, and V . Kumar, Anomaly Detection: A Survey, ACM Comput- ing Surveys (CSUR), 41(3), 2009, 1–58

    2009

  10. [18]

    J. P. S. Chhabra and G. P. Warn, A Method for Model Selection Using Reinforcement Learn- ing When Viewing Design as a Sequential Decision Process, Structural and Multidisci- plinary Optimization, 59, 2019, 1521–1542

    2019

  11. [19]

    Sutton and A.G

    R.S. Sutton and A.G. Barto, Reinforcement Learning, second edition: An Introduction, MIT Press, 2018

    2018

  12. [20]

    Kosana, K

    V . Kosana, K. Teeparthi, S. Madasthu, and S. Kumar, A Novel Reinforced Online Model Se- lection Using Q-learning Technique for Wind Speed Prediction, Sustainable Energy Tech- nologies and Assessments, 49, 2022, 101780

    2022

  13. [21]

    Y . Fu, D. Wu, and B. Boulet, Reinforcement Learning Based Dynamic Model Combination for Time Series Forecasting, Proceedings of the AAAI Conference on Artificial Intelligence, 36(6), 2022, 6639–6647

    2022

  14. [22]

    Christophe, J

    K. Christophe, J. El Zini, and M. Awad, A Progressive and Cross-Domain Deep Transfer Learning Framework for Wrist Fracture Detec- tion, Journal of Artificial Intelligence and Soft Computing Research, 12(2), 2021, 101-120

    2021

  15. [23]

    H. Deng, G. Runger, E. Tuv, and V . Martyanov, A Time Series Forest for Classification and Feature Extraction, Information Sciences, 239, 2013, 142–153

    2013

  16. [24]

    Philosophical Transactions of the Royal Soci- ety A, 379(2194), 2021, 20200209

    Lim, Bryan, Zohren, and Stefan, Time-Series Forecasting with Deep Learning: A Survey. Philosophical Transactions of the Royal Soci- ety A, 379(2194), 2021, 20200209

    2021

  17. [25]

    Braei and S

    M. Braei and S. Wagner, Anomaly Detection in Univariate Time-Series: A Survey on the State- of-the-Art, arXiv preprint arXiv:2004.00433, 2020

    2004 arXiv

  18. [26]

    Aggarwal, Outlier Analysis, Springer In- ternational Publishing, 2016

    C.C. Aggarwal, Outlier Analysis, Springer In- ternational Publishing, 2016

    2016

  19. [27]

    G ¨ornitz, M

    N. G ¨ornitz, M. Kloft, K. Rieck, and U. Brefeld, Toward Supervised Anomaly Detection, Jour- nal of Artificial Intelligence Research, 46, 2013, 235–262

    2013

  20. [28]

    V . N. Vapnik, An Overview of Statistical Learning Theory, IEEE Transactions on Neural Networks, 10(5), 1999, 988–999

    1999

  21. [29]

    Sindhwani, P

    V . Sindhwani, P. Niyogi, and M. Belkin, Be- yond the Point Cloud: From Transductive to Semi-Supervised Learning, Proceedings of the 22nd International Conference on Machine Learning, 2005, 824–831

    2005

  22. [30]

    E. M. Knorr, R. T. Ng, and V . Tucakov, Distance-Based Outliers: Algorithms and Ap- plications, The VLDB Journal, 8(3), 2000, 237–253. 21

    2000

  23. [31]

    Ramaswamy, R

    S. Ramaswamy, R. Rastogi, and K. Shim, Ef- ficient Algorithms for Mining Outliers from Large Data Sets, Proceedings of the 2000 ACM SIGMOD International Conference on Man- agement of Data, 2000, 427–438

    2000

  24. [32]

    Angiulli and C

    F. Angiulli and C. Pizzuti, Fast Outlier De- tection in High Dimensional Spaces, European Conference on Principles of Data Mining and Knowledge Discovery, 2002, 15–27

    2002

  25. [33]

    M. M. Breunig, H.-P. Kriegel, R. T. Ng, and J. Sander, LOF: Identifying Density-Based Lo- cal Outliers, Proceedings of the 2000 ACM SIGMOD International Conference on Man- agement of Data, 2000, 93–104

    2000

  26. [34]

    G. O. Campos, A. Zimek, J. Sander, R. J. G. B. Campello, B. Micenkov ´a, E. Schubert, I. Assent, and M. E. Houle, On the Evaluation of Unsupervised Outlier Detection: Measures, Datasets, and an Empirical Study, Data Mining and Knowledge Discovery, 30, 2016, 891–927

    2016

  27. [35]

    Xiong, X

    L. Xiong, X. Chen, and J. Schneider, Direct Ro- bust Matrix Factorization for Anomaly Detec- tion, 2011 IEEE 11th International Conference on Data Mining, 2011, 844–853

    2011

  28. [36]

    L. Li, J. McCann, N. S. Pollard, and C. Falout- sos, Dynammo: Mining and Summarization of Coevolving Sequences with Missing Values, Proceedings of the 15th ACM SIGKDD Inter- national Conference on Knowledge Discovery and Data Mining, 2009, 507–516

    2009

  29. [37]

    G ¨ornitz, M

    N. G ¨ornitz, M. Braun, and M. Kloft, Hid- den Markov Anomaly Detection, International Conference on Machine Learning, 2015, 1833– 1842

    2015

  30. [38]

    A. P. Dawid and A. M. Skene, Maximum Like- lihood Estimation of Observer Error-Rates Us- ing the EM Algorithm, Journal of the Royal Statistical Society: Series C (Applied Statis- tics), 28(1), 1979, 20–28

    1979

  31. [39]

    Awad, and R

    M. Awad, and R. Khanna, Efficient Learn- ing Machines: Theories, Concepts, and Appli- cations for Engineers and System Designers, Springer Nature, 2015

    2015

  32. [40]

    L. M. Manevitz and M. Yousef, One-Class SVMs for Document Classification, Journal of Machine Learning Research, 2(Dec), 2001, 139–154

    2001

  33. [41]

    Sch ¨olkopf, J

    B. Sch ¨olkopf, J. C. Platt, J. Shawe-Taylor, A. J. Smola, and R. C. Williamson, Estimating the Support of a High-Dimensional Distribution, Neural Computation, 13(7), 2001, 1443–1471

    2001

  34. [42]

    Z. Li, Y . Zhao, X. Hu, N. Botta, C. Ionescu, and G. H. Chen, ECOD: Unsupervised Outlier De- tection Using Empirical Cumulative Distribu- tion Functions, IEEE Transactions on Knowl- edge and Data Engineering, 35(12), 2022, 12181–12193

    2022

  35. [43]

    Z. Li, Y . Zhao, N. Botta, C. Ionescu, and X. Hu, COPOD: Copula-Based Outlier Detection, 2020 IEEE International Conference on Data Mining (ICDM), 2020, 1118–1123

    2020

  36. [44]

    Kharitonov, A

    A. Kharitonov, A. Nahhas, M. Pohl, and K. Turowski, Comparative Analysis of Machine Learning Models for Anomaly Detection in Manufacturing, Procedia Computer Science, 200, 2022, 1288–1297

    2022

  37. [45]

    Z. Xu, D. Kakde, and A. Chaudhuri, Automatic Hyperparameter Tuning Method for Local Out- lier Factor, with Applications to Anomaly De- 22 tection, 2019 IEEE International Conference on Big Data (Big Data), 2019, 4201–4207

    2019

  38. [46]

    F. T. Liu, K. M. Ting, and Z.-H. Zhou, Isolation Forest, 2008 Eighth IEEE International Confer- ence on Data Mining, 2008, 413–422

    2008

  39. [47]

    P. J. Rousseeuw and M. Hubert, Anomaly De- tection by Robust Statistics, Wiley Interdisci- plinary Reviews: Data Mining and Knowledge Discovery, 8(2), 2018, e1236

    2018

  40. [48]

    Hoffmann, Kernel PCA for Novelty Detec- tion, Pattern Recognition, 40(3), 2007, 863– 874

    H. Hoffmann, Kernel PCA for Novelty Detec- tion, Pattern Recognition, 40(3), 2007, 863– 874

    2007

  41. [49]

    Malhotra, A

    P. Malhotra, A. Ramakrishnan, G. Anand, L. Vig, P. Agarwal, and G. Shroff, LSTM-Based Encoder-Decoder for Multi-Sensor Anomaly Detection, arXiv preprint arXiv:1607.00148, 2016

    2016 arXiv

  42. [50]

    Y . Su, Y . Zhao, C. Niu, R. Liu, W. Sun, and D. Pei, Robust Anomaly Detection for Multivari- ate Time Series through Stochastic Recurrent Neural Network, Proceedings of the 25th ACM SIGKDD International Conference on Knowl- edge Discovery & Data Mining, 2019, 2828– 2837

    2019

  43. [51]

    Hochreiter and J

    S. Hochreiter and J. Schmidhuber, Long Short- Term Memory, Neural Computation, 9(8), 1997, 1735–1780

    1997

  44. [52]

    El Zini, Y

    J. El Zini, Y . Rizk, and M. Awad, An Optimized Parallel Implementation of Non- Iteratively Trained Recurrent Neural Networks, Journal of Artificial Intelligence and Soft Com- puting Research, 11(1), 2021, 33-50

    2021

  45. [53]

    Audibert, P

    J. Audibert, P. Michiardi, F. Guyard, S. Marti, and M. A. Zuluaga, USAD: Unsupervised Anomaly Detection on Multivariate Time Se- ries, Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Dis- covery & Data Mining, 2020, 3395–3404

    2020

  46. [54]

    Gonz ´alez-Carvajal and E

    S. Gonz ´alez-Carvajal and E. C. Garrido- Merch´an, Comparing BERT Against Tradi- tional Machine Learning Text Classification, arXiv preprint arXiv:2005.13012, 2020

    2005 arXiv

  47. [55]

    W. Dang, B. Zhou, W. Zhang, and S. Hu, Time Series Anomaly Detection Based on Language Model, Proceedings of the Eleventh ACM In- ternational Conference on Future Energy Sys- tems, 2020, 544–547

    2020

  48. [56]

    M. Dong, H. Huang, and L. Cao, Can LLMs Serve As Time Series Anomaly Detectors?, arXiv preprint arXiv:2408.03475, 2024

    2024 arXiv

  49. [57]

    J. Su, C. Jiang, X. Jin, Y . Qiao, T. Xiao, H. Ma, R. Wei, Z. Jing, J. Xu, and J. Lin, Large Lan- guage Models for Forecasting and Anomaly Detection: A Systematic Literature Review, arXiv preprint arXiv:2402.10350, 2024

    2024 arXiv

  50. [58]

    Y . Li, Z. Chen, D. Zha, K. Zhou, H. Jin, H. Chen, and X. Hu, AUTOOD: Neural Architec- ture Search for Outlier Detection, 2021 IEEE 37th International Conference on Data Engi- neering (ICDE), 2021, 2117–2122

    2021

  51. [59]

    K.-H. Lai, D. Zha, G. Wang, J. Xu, Y . Zhao, D. Kumar, Y . Chen, P. Zumkhawaka, M. Wan, D. Martinez, et al., TODS: An Automated Time Series Outlier Detection System, Proceedings of the AAAI Conference on Artificial Intelli- gence, 35(18), 2021, 16060–16062

    2021

  52. [60]

    Y . Li, D. Zha, P. Venugopal, N. Zou, and X. Hu, PYODDS: An End-to-End Outlier Detec- tion System with Automated Machine Learn- 23 ing, Companion Proceedings of the Web Con- ference 2020, 2020, 153–157

    2020

  53. [61]

    Y . Zhao, R. Rossi, and L. Akoglu, Automatic Unsupervised Outlier Model Selection, Ad- vances in Neural Information Processing Sys- tems, 34, 2021, 4489–4502

    2021

  54. [62]

    Gulati and P

    M. Gulati and P. Arjunan, LEAD1.0: A Large- Scale Annotated Dataset for Energy Anomaly Detection in Commercial Buildings, Proceed- ings of the Thirteenth ACM International Con- ference on Future Energy Systems, 2022, 485– 488

    2022

  55. [63]

    A Platform for Open Data of the Euro- pean Power System, available at https:// open-power-system-data.org/ , Accessed on: Aug. 9, 2024

    2024

  56. [64]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, et al., Scikit-Learn: Machine Learning in Python, The Journal of Machine Learning Research, 12, 2011, 2825–2830

    2011

  57. [65]

    Y . Zhao, Z. Nasrullah, and Z. Li, PYOD: A Python Toolbox for Scalable Outlier Detection, Journal of Machine Learning Research, 20(96), 2019, 1–7

    2019

  58. [66]

    Galati, USAD: UnSupervised Anomaly De- tection on Multivariate Time Series, available at https://github.com/manigalati/usad,

    M. Galati, USAD: UnSupervised Anomaly De- tection on Multivariate Time Series, available at https://github.com/manigalati/usad,

  59. [67]

    Bergstra, B

    J. Bergstra, B. Komer, C. Eliasmith, D. Yamins, and D. D. Cox, Hyperopt: A Python Library for Model Selection and Hyperparameter Opti- mization, Computational Science & Discovery, 8(1), 2015, 014008

    2015

  60. [68]

    Raffin, A

    A. Raffin, A. Hill, A. Gleave, A. Kan- ervisto, M. Ernestus, and N. Dormann, Stable- Baselines3: Reliable Reinforcement Learning Implementations, Journal of Machine Learning Research, 22(268), 2021, 1–8. A Hyperparameter Choices Hyperparameter 1 Hyperparameter 2 n neighbors: [...

    2021

  61. [2020]

    Accessed on: Aug. 9, 2024

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.