Calibrated Unsupervised Anomaly Detection in Multivariate Time-series using Reinforcement Learning

Amir G. Aghdam; Saba Sanami

REVIEW 4 major objections 4 minor 20 references

Calibrated Unsupervised Anomaly Detection in Multivariate Time-series using Reinforcement Learning

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read A reinforcement-learning agent that calibrates the decision boundary inside an autoencoder's latent space is claimed to lift unsupervised anomaly-detection F1 to 0.931 on the C-MAPSS aero-engine dataset.

desk verdict The pipeline combines known parts in a new way, but the evaluation is circular: the test labels are thresholded versions of the same scores the detector uses, so Table II doesn't demonstrate real anomaly detection. read the letter →

arxiv 2502.03245 v2 pith:PE4SXUZN submitted 2025-02-05 cs.LG cs.SYeess.SPeess.SY

Saba Sanami , Amir G. Aghdam This is my paper · ORCID

classification cs.LGcs.SYeess.SPeess.SY

keywords unsupervisedanomalydetectionmultivariatetimeseriesreinforcementlearningQ-learningautoencoderwavelettransformdecisionboundarycalibrationC-MAPSS

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

The paper tries to establish that unsupervised anomaly detection in multivariate time series can be improved by letting a reinforcement learning agent calibrate the detector's decision boundary inside an autoencoder's latent space. The authors argue that scarce anomaly labels inflate false negatives, and that synthetic anomalies plus RL exploration can compensate. On the C-MAPSS aero-engine dataset, the method is reported to reach precision 0.921, recall 0.942, accuracy 0.911, and F1 0.931, beating an LSTM-AE baseline on every metric. A sympathetic reader would care because the approach addresses a practical setting where true anomaly labels are unavailable.

What carries the argument

The load-bearing mechanism is an RL loop around the latent space. The state is the latent code $z_n = f_{\text{enc}}(w_n; \theta_{\text{enc}})$; the action is $a_n = (\Delta\theta_{\text{enc}}, \Delta\theta_n)$; and the reward is $R(s_n, a_n) = R_{\text{sep}} + R_{\text{acc}}$, where $R_{\text{sep}}$ is the squared distance between normal and synthetic-anomaly centroids and $R_{\text{acc}}$ is +1 or -1 for correct or incorrect classification under the current threshold. Training minimizes $L_{\text{total}} = L_{\text{recon}} - \eta Q(s_n, a_n)$, so reconstruction fidelity and expected future reward pull in the same direction. Synthetic anomalies injected into training provide the supervised reference points that let the agent calibrate the boundary.

What would settle it

Re-run the comparison on C-MAPSS using the actual degradation labels (for instance, windows after the known failure-progression threshold) instead of pseudo-labels derived from reconstruction error and uncertainty; if the proposed method's precision and recall drop to the baseline's level, the reported superiority was an artifact of circular labeling.

Watch

Extended reading notes

Core claim

The central claim is that the combination of wavelet-based multiresolution features, a convolutional autoencoder, and an RL-calibrated boundary yields a better-performing unsupervised anomaly detector. The discrete wavelet transform turns each sliding window into an image-like coefficient stack; the CNN autoencoder learns a three-dimensional latent space; and a Q-learning agent adjusts both encoder parameters and the threshold so that normal data and synthetic anomalies form separated clusters. The paper reports that this pipeline outperforms LSTM-AE on all metrics, with F1 0.931 versus 0.802.

Load-bearing premise

The evaluation assumes that points with high reconstruction error and high latent uncertainty really are the anomalies, so the ground truth used to compute precision, recall, and F1 is built from the same signal the detector uses to decide; if that labeling is wrong, the reported advantage over LSTM-AE is not grounded.

Editorial extensions

If this is right

If the reported numbers hold, the same three-stage recipe — wavelet features, CNN autoencoder, and RL threshold calibration — should transfer to other multivariate monitoring tasks where anomalies are rare.
A calibrated boundary implies fewer false negatives, which matters in safety-critical monitoring such as engine health.
Synthetic anomaly injection becomes a viable way to supervise the boundary without real labels, meaning practitioners can design anomaly types for their own failure modes.
The RL component gives the detector an adaptive threshold that can track changes in the latent-space distribution over time.

Reading between the lines

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

Editorial inference: the reported evaluation is circular in the sense the paper itself describes, because the pseudo-labels come from high reconstruction error and high uncertainty — exactly the anomaly scores the detector uses — so a fair comparison would need independent labels.
Editorial inference: the RL agent as described adjusts a threshold and encoder gradients, so the improvement over LSTM-AE could come chiefly from the wavelet inputs or the CNN architecture rather than from RL per se.
Editorial inference: a direct test would replace the Q-learning update with a simple percentile threshold on reconstruction error; if the RL gain disappears, boundary calibration is not the active ingredient.
Editorial inference: the framework suggests a broader recipe — synthetic anomaly injection plus an explicit separation reward could calibrate any reconstruction-based detector, not just this autoencoder.

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.

Desk Editor's Note

read the letter

The paper's novelty is combinatorial: wavelet coefficients fed to a CNN autoencoder, with a Q-learning agent adjusting the decision boundary and synthetic anomalies injected to guide calibration. That specific pipeline isn't in the cited literature, and the paper explains it clearly enough to reproduce the architecture. The latent-space figure is a nice touch, and the wavelet preprocessing is a reasonable idea.

The problem is the evaluation, and it's a load-bearing one. In Section IV, the authors state that true labels aren't available, so they "devise a labeling strategy": points with high reconstruction error and high uncertainty are called anomalies, and low error plus low uncertainty are called normal. But the detector's own rule is exactly that—Algorithm 1 labels a point anomalous when reconstruction error exceeds an adjusted threshold, and the uncertainty grouping comes from Monte Carlo dropout in the same latent space. So the precision, recall, accuracy, and F1 in Table II measure how well a threshold on the reconstruction error matches a label set that is itself a threshold on that same error. That's circular, and it means the reported outperformance over LSTM-AE says nothing about detecting real anomalies in C-MAPSS. The dataset has run-to-failure trajectories with remaining-useful-life annotations that could supply external labels; the paper doesn't use them. It also counts synthetic anomalies as test anomalies even though they were used during training, which inflates the numbers.

There are secondary issues. The RL formalism is inconsistent: Eq. (5) says the action updates both encoder parameters and the boundary, but Algorithm 1 only updates the boundary, and line 15 sets the next state to the current state. The reward is a classification reward on synthetic labels, so the "reinforcement learning" reduces to a threshold search. There are no error bars, no ablations, and the only baseline is a plain LSTM-AE, which is not a competitive modern comparison.

I agree with the stress-test: this is not a manufactured flaw. The circular evaluation breaks the central claim. The ingredients—wavelet features and synthetic-anomaly calibration—are worth keeping, but they need to be validated against real anomaly labels, and the RL part needs to actually do what the equations say.

I wouldn't send this to referees as is. A desk reject with an invitation to resubmit after a proper C-MAPSS evaluation (using RUL-based labels, no synthetic anomalies in the test set) and a cleaned-up RL formulation would be the right call. As presented, the paper doesn't justify referee time.

Referee Report

4 major / 4 minor

Summary. The paper proposes an unsupervised anomaly detection framework for multivariate time-series data that combines discrete wavelet transform (DWT) feature extraction, a convolutional neural network autoencoder (CNN-AE), and a Q-learning-based agent intended to calibrate the decision boundary using synthetic anomalies. The method is evaluated on the C-MAPSS aero-engine dataset and compared against an LSTM-AE baseline. The central empirical claim, stated in Section IV, is that the proposed method outperforms LSTM-AE on precision, recall, accuracy, and F1-score. However, the evaluation labels are constructed from the model's own reconstruction-error and uncertainty scores, making the reported metrics circular; additionally, the RL formalism is internally inconsistent between the equations and the algorithm. The paper concludes by suggesting future work with DQN or policy gradient methods.

Significance. If the empirical claim were valid, the integration of wavelet-based features, uncertainty estimation via Monte Carlo dropout, and RL-driven decision-boundary calibration could offer a meaningful contribution to unsupervised anomaly detection in multivariate time-series. The paper is clearly written in terms of structure, and it addresses a real challenge: the scarcity of anomalous labels. Credit is due for attempting to incorporate uncertainty and synthetic anomalies into boundary calibration on a standard benchmark dataset (C-MAPSS). However, the central evaluation is circular, and the RL component as described is not a well-defined algorithm. Because the claimed superiority over LSTM-AE rests on these defective pillars, the paper does not currently establish its stated contribution.

major comments (4)

[Section IV, 'We employ several evaluation metrics' paragraph; Table II] The evaluation is circular: the ground-truth labels used to compute precision, recall, accuracy, and F1 are defined by thresholding the same scores the model uses for detection. The paragraph states: 'instances with synthetic anomalies, high reconstruction errors, and high uncertainty in the latent space are treated as anomalies... Conversely, data with low reconstruction error and low uncertainty are considered normal.' Algorithm 1 (lines 11-13) predicts a label based on whether the reconstruction error e_n exceeds the adjusted threshold θ_n, and the uncertainty grouping is also derived from the model's latent-space representations. Thus, the reported metrics in Table II measure agreement between two thresholds on the same model outputs, not detection of actual anomalies. The comparison with LSTM-AE does not resolve this, because the labels are constructed from the proposed model's scores, not from independent ground truth. Furthermore, synthetic anomalies are included in the test labels even though they were generated and used during training, which constitutes data leakage. C-MAPSS provides run-to-failure trajectories from which anomaly labels could be derived (e.g., segments near the end of life), but the paper does not use them; the authors also explicitly acknowledge 'true labels for the anomalies are not available.' This invalidates the central empirical claim that the proposed method outperforms LSTM-AE.
[Section III-B, Eqs. (5)-(6), Eq. (10), Algorithm 1] The RL formalism is internally inconsistent. Eq. (5) defines the action as a tuple of continuous adjustments, a_n = (Δθ_enc, Δθ_n), and Eq. (6) defines the next state as s_{n+1} = f_enc(w_{n+1}; θ_enc + Δθ_enc). However, Algorithm 1 line 10 applies only θ_n ← θ_n + Δθ_n, and line 15 sets s_{n+1} = s_n, ignoring the encoder update and the next-state equation. Line 19 updates the encoder parameters by minimizing L_total, not by the RL action. Moreover, Section IV states 'two possible actions for the RL agent, classify data as normal or classify it as abnormal,' which is a discrete classification action, not the continuous boundary/encoder adjustment of Eq. (5). Tabular Q-learning is then applied to a state space consisting of continuous latent representations, for which tabular Q-learning is not well defined. As written, the RL component is not a concrete, reproducible algorithm, and the claimed mechanism of dynamic decision-boundary calibration via RL is not actually implemented.
[Section III-B, Eq. (8); Section IV, synthetic anomaly generation] The reward function in Eq. (8) uses labels for synthetic anomalies and assumes all other data are normal (y_n = 0). This is a supervised signal, contradicting the paper's 'unsupervised anomaly detection' framing. More importantly, the synthetic anomalies are generated by perturbing normal signals (cyclic, sudden drift, gradual drift), and these same anomalies are used both to train the RL agent and to define part of the test labels. The paper provides no evidence that these synthetic perturbations are representative of real C-MAPSS degradation anomalies; indeed, Section IV states 'the exact point of abnormality is uncertain.' The calibrated decision boundary may therefore be tuned to artifacts of the perturbation method rather than to actual faults, and the reported performance on synthetic anomalies does not transfer to real anomalies. This is a load-bearing assumption that is not validated.
[Section IV, experimental setup and Table II] The experimental comparison lacks statistical rigor: no standard deviations, no multiple independent runs, no hyperparameter sensitivity analysis, and only a single baseline (LSTM-AE). The reported differences (e.g., F1 0.931 vs. 0.802) could be within random variation, especially given the small evaluation set and the circular labeling. Without error bars or a significance test, the claim of 'consistently outperformed' is not supported. Even if the circularity were resolved, the paper would need additional baselines and repeated trials to demonstrate a reliable advantage.

minor comments (4)

[Title page] There is a typo in the affiliation: 'Concordia Univerity' should be 'Concordia University.'
[Section III-B, Eq. (11) and Algorithm 1 line 19] Eq. (11) defines L_total = L_recon - η Q(s_n, a_n), but Algorithm 1 line 19 sums over all n, Σ_n Q(s_n, a_n). The notation should be harmonized to avoid ambiguity about whether the Q-term is a per-sample or batch term.
[Section IV, Fig. 5 description] The figure caption text says 'yellow circles represent real abnormal data points that were classified as high uncertainty by the model.' Since no ground-truth anomaly labels are used, these points should be described as 'high-uncertainty data points,' not 'real abnormal data points,' to avoid implying an external ground truth.
[Section IV, Table I and text] Table I lists 'Number of episode' (should be 'episodes'); also, the text says 'We initialize a Q-table with zeros,' but the state and action spaces as described in Section III-B are continuous, so a Q-table would be infinite. This inconsistency should be clarified.

Circularity Check

3 steps flagged · score 8.0 of 10

Section IV's evaluation is circular: test labels are thresholds on the model's own reconstruction error (Eq. 3) and MC-dropout uncertainty, the same scores used as the detector, so Table II reports self-consistency rather than detection of actual C-MAPSS anomalies.

self definitional [Section IV, evaluation-metrics paragraph; cf. Eq. (3) and Eq. (8)]
"Specifically, instances with synthetic anomalies, high reconstruction errors, and high uncertainty in the latent space are treated as anomalies, as these often indicate deviation from learned normal patterns. Conversely, data with low reconstruction error and low uncertainty are considered normal."

The detector's positive class is defined in Eq. (8) and Algorithm 1 (lines 11-12) as y_hat_n = 1 exactly when e_n > theta_n, with e_n = ||w_n - w_hat_n||^2 (Eq. 3). The evaluation labels test points as anomalous exactly when the same e_n is high and the same latent-space MC-dropout uncertainty is high. A threshold rule on e_n therefore trivially agrees with labels that are thresholds on e_n and on the model's own uncertainty, so the precision/recall/F1 values in Table II measure self-consistency, not detection of true C-MAPSS anomalies. Synthetic anomalies, generated during training (Algorithm 1 line 3), are also included in the positive test set.
fitted input called prediction [Section IV, paragraph describing Fig. 5]
"The yellow circles represent real abnormal data points that were classified as high uncertainty by the model."

The high-uncertainty category is not an external label: the paper states that data points are placed into low/high uncertainty groups 'based on a threshold set at the 75th percentile of uncertainty scores' (Section IV). Thus 'real abnormal data' is by construction whatever the model's MC-dropout uncertainty places in the top quartile. The subsequent claim that the RL boundary 'encompasses both synthetic and real abnormal data points' and the quantitative results in Table II are therefore evaluated against labels generated by the model's own uncertainty estimates, not against independently known C-MAPSS fault labels.

1 more flagged steps

fitted input called prediction [Section IV, Table II comparison]
"We then compare the corresponding metrics with those of LSTM-AE, as shown in Table II. The proposed method consistently outperformed the LSTM-AE model in all metrics, confirming its superior performance in anomaly detection."

The comparison uses the same self-generated labels. Because the labels are thresholds on the proposed model's reconstruction error and uncertainty, the proposed model's predictions are matched to its own definitions, while LSTM-AE is judged against a label set constructed from the proposed model's scores. The reported outperformance in Table II is therefore an artifact of the labeling strategy rather than evidence that the proposed method detects real anomalies better on C-MAPSS.

full rationale

The proposed architecture itself (wavelet features, CNN autoencoder, and RL boundary calibration) is an independent construction, and I found no load-bearing self-citation chain: reference [16] is a background citation and not used to justify the central result. However, the empirical evaluation is circular in a way that invalidates the paper's strongest claim. The test labels are defined in Section IV as points with high reconstruction error and high uncertainty in the latent space, which are precisely the model's own anomaly scores (Eq. 3 and the MC-dropout uncertainty threshold). Since the model's predicted label is y_hat_n = 1 if e_n > theta_n (Eq. 8, Algorithm 1), any threshold-based detector will trivially score well against labels that are thresholds on the same e_n and on the model's own uncertainty output. Consequently, precision, recall, accuracy, and F1 in Table II, and the claimed outperformance over LSTM-AE, measure self-consistency rather than detection of true C-MAPSS anomalies. The paper admits it creates synthetic abnormalities and uses them as labels, and it never uses genuine run-to-failure or RUL annotations as test labels. The central empirical claim therefore reduces by construction to the model's own score definitions, warranting a high circularity score of 8.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several hand-chosen thresholds and rewards, on the unvalidated assumption that synthetic anomalies resemble real faults, and on an evaluation scheme whose labels are generated by the same reconstruction-error and uncertainty scores the model predicts. The unspecified eta and synthetic anomaly parameters further limit the ability to audit the method.

free parameters (6)

Initial decision boundary theta_0 = 0.5
Hand-set in Table I; the whole calibration exercise is about adjusting this boundary, so its initial value materially affects the reported behavior.
Uncertainty group threshold = 75th percentile of uncertainty scores
Chosen by hand in Section IV; points above this percentile are treated as high uncertainty and therefore abnormal in the evaluation.
Reward weight eta in L_total = not specified
Eq. (11) balances reconstruction loss and Q-value with an unspecified weight; no value is given in Table I or the text, making the objective incompletely defined.
Reward weight for separation lambda = 1.0
Given in Table I; controls the contribution of the separation reward, but the separation reward is not actually computed in Algorithm 1.
Synthetic anomaly generation parameters = not specified
The paper describes three anomaly types but gives no amplitudes, durations, counts, or injection rules; these choices determine the calibration signal and the evaluation set.
RL optimizer hyperparameters = alpha=0.01, gamma=0.95, epsilon=0.1, decay=0.99, episodes=1000
Listed in Table I; standard hyperparameters but chosen by hand and not varied, so their influence on the central claim is untested.

assumptions (4)

domain assumption Reconstruction error and Monte Carlo dropout uncertainty are jointly sufficient statistics for anomaly presence.
The detector and the evaluation both rely on this assumption. It is introduced in Section III-A and used in the evaluation paragraph of Section IV.
ad hoc to paper Synthetic anomalies created by perturbing normal signals are representative of real anomalies in C-MAPSS.
The whole calibration scheme depends on this assumption, stated in Section III-B and the experimental setup, but no evidence connects the synthetic perturbations to actual engine degradation signatures.
ad hoc to paper A tabular Q-learning agent can optimize continuous encoder updates and a continuous decision boundary from latent states.
Eq. (5) defines actions as (Delta_theta_enc, Delta_theta_n), but Algorithm 1 implements only two discrete classification actions and sets the next state equal to the current state, so the formalism does not match the implementation.
domain assumption C-MAPSS provides no usable true anomaly labels for evaluation.
The paper states true labels are not available, yet C-MAPSS provides remaining useful life and fault condition information from which degradation segments could be derived. This assumption justifies the circular labeling scheme.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Calibrated Unsupervised Anomaly Detection in Multivariate Time-series using Reinforcement Learning." pith.science (2026). https://pith.science/paper/PE4SXUZN

@misc{pith2026250203245,
  author       = {Pith},
  title        = {Pith review of: Calibrated Unsupervised Anomaly Detection in Multivariate Time-series using Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PE4SXUZN}},
  note         = {Machine review of arXiv:2502.03245}
}

read the original abstract

This paper investigates unsupervised anomaly detection in multivariate time-series data using reinforcement learning (RL) in the latent space of an autoencoder. A significant challenge is the limited availability of anomalous data, often leading to misclassifying anomalies as normal events, thus raising false negatives. RL can help overcome this limitation by promoting exploration and balancing exploitation during training, effectively preventing overfitting. Wavelet analysis is also utilized to enhance anomaly detection, enabling time-series data decomposition into both time and frequency domains. This approach captures anomalies at multiple resolutions, with wavelet coefficients extracted to detect both sudden and subtle shifts in the data, thereby refining the anomaly detection process. We calibrate the decision boundary by generating synthetic anomalies and embedding a supervised framework within the model. This supervised element aids the unsupervised learning process by fine-tuning the decision boundary and increasing the model's capacity to distinguish between normal and anomalous patterns effectively.

Figures

Figures reproduced from arXiv: 2502.03245 by the authors.

**Figure 2.** DWT coefficients for first and last 5 sequences [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** HPC outlet temperature with synthetic anomalies [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: DWT coefficient for synthetic anomalies sequences [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: Latent space representation with RL decision boundary [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 6.** Figure 6: represents the cumulative reward for Q-learning training over 1000 episodes. Initially, the agent experiences some negative rewards as it explores its environment and tries different actions without much knowledge. However, as training progresses, we observe a signific…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 19 canonical work pages

[1]

Anomaly detection on time series,

M. Teng, “Anomaly detection on time series,” in proceeding of IEEE International Conference on Progress in Informatics and Computing , vol. 1, 2010, pp. 603–608

work page 2010
[2]

Deep learning for anomaly detection in time-series data: Review, analysis, and guidelines,

K. Choi, J. Yi, C. Park, and S. Yoon, “Deep learning for anomaly detection in time-series data: Review, analysis, and guidelines,” IEEE Access, vol. 9, pp. 120 043–120 065, 2021

2021
[3]

Outlier detection and missing value in seasonal ARIMA model using rainfall data,

P. Arumugam and R. Saranya, “Outlier detection and missing value in seasonal ARIMA model using rainfall data,” Materials Today: Proceed- ings, vol. 5, no. 1, pp. 1791–1799, 2018

work page 2018
[4]

Modeling data containing outliers using ARIMA additive outlier (ARIMA-AO),

A. S. Ahmar, S. Guritno, A. Rahman, I. Minggi, M. A. Tiro, M. K. Aidid, S. Annas, D. U. Sutiksno, D. S. Ahmar, K. H. Ahmar et al., “Modeling data containing outliers using ARIMA additive outlier (ARIMA-AO),” in Journal of Physics: Conference Series, vol. 954, no. 1. IOP Publishing, 2018, p. 012010

work page 2018
[5]

A machine learning framework for network anomaly detection using SVM and GA,

T. Shon, Y . Kim, C. Lee, and J. Moon, “A machine learning framework for network anomaly detection using SVM and GA,” in Proceedings from the 6th annual IEEE SMC information assurance workshop. IEEE, 2005, pp. 176–183

work page 2005
[6]

Anomaly detection based on machine learning: dimensionality reduction using PCA and classification using SVM,

A. George and A. Vidyapeetham, “Anomaly detection based on machine learning: dimensionality reduction using PCA and classification using SVM,” International Journal of Computer Applications , vol. 47, no. 21, pp. 5–8, 2012

work page 2012
[7]

A novel anomaly detection method based on adaptive mahalanobis-squared distance and one-class kNN rule for structural health monitoring under environmental effects,

H. Sarmadi and A. Karamodin, “A novel anomaly detection method based on adaptive mahalanobis-squared distance and one-class kNN rule for structural health monitoring under environmental effects,” Mechan- ical Systems and Signal Processing , vol. 140, p. 106495, 2020

work page 2020
[8]

Unsupervised anomaly detection in time series using LSTM-based autoencoders,

O. I. Provotar, Y . M. Linder, and M. M. Veres, “Unsupervised anomaly detection in time series using LSTM-based autoencoders,” in 2019 IEEE International Conference on Advanced Trends in Information Theory (ATIT), 2019, pp. 513–517

work page 2019

Show all 20 references

[9]

Long short term memory networks for anomaly detection in time series

P. Malhotra, L. Vig, G. Shroff, P. Agarwal et al. , “Long short term memory networks for anomaly detection in time series.” p. 89, 2015

2015
[10]

Unsupervised anomaly detection with LSTM neural networks,

T. Ergen and S. S. Kozat, “Unsupervised anomaly detection with LSTM neural networks,” IEEE Transactions on Neural Networks and Learning Systems, vol. 31, no. 8, pp. 3127–3141, 2019

2019
[11]

Improved LSTM-based time-series anomaly detection in rail transit operation environments,

Y . Wang, X. Du, Z. Lu, Q. Duan, and J. Wu, “Improved LSTM-based time-series anomaly detection in rail transit operation environments,” IEEE Transactions on Industrial Informatics , vol. 18, no. 12, pp. 9027– 9036, 2022

2022
[12]

Improving lead time forecasting and anomaly detection for automotive spare parts with a combined CNN-LSTM approach,

A. Amellal, I. Amellal, H. Seghiouer, and M. Ech-Charrat, “Improving lead time forecasting and anomaly detection for automotive spare parts with a combined CNN-LSTM approach,” Operations and Supply Chain Management: An International Journal , vol. 16, no. 2, pp. 265–278, 2023

2023
[13]

Time-series anomaly detection with stacked transformer representations and 1D convolutional network,

J. Kim, H. Kang, and P. Kang, “Time-series anomaly detection with stacked transformer representations and 1D convolutional network,” Engineering Applications of Artificial Intelligence , vol. 120, p. 105964, 2023

2023
[14]

Unsupervised anomaly detection

D. Guthrie, L. Guthrie, B. Allison, and Y . Wilks, “Unsupervised anomaly detection.” in proceeding of International Joint Conference on Artificial Intelligence, 2007, pp. 1624–1628

2007
[15]

Unsupervised anomaly detection in time-series: An exten- sive evaluation and analysis of state-of-the-art methods,

N. Mejri, L. Lopez-Fuentes, K. Roy, P. Chernakov, E. Ghorbel, and D. Aouada, “Unsupervised anomaly detection in time-series: An exten- sive evaluation and analysis of state-of-the-art methods,” Expert Systems with Applications, p. 124922, 2024

2024
[16]

Aero-engines anomaly detection using an unsupervised Fisher autoencoder,

S. Sanami and A. G. Aghdam, “Aero-engines anomaly detection using an unsupervised Fisher autoencoder,” in proceeding of 63rd IEEE Conference on Decision and Control , 2024

2024
[17]

Calibrated one-class classification for unsupervised time series anomaly detection,

H. Xu, Y . Wang, S. Jian, Q. Liao, Y . Wang, and G. Pang, “Calibrated one-class classification for unsupervised time series anomaly detection,” IEEE Transactions on Knowledge and Data Engineering , 2024

2024
[18]

A hybrid model based on discrete wavelet transform (DWT) and bidirectional recurrent neural networks for wind speed prediction,

A. Barjasteh, S. H. Ghafouri, and M. Hashemi, “A hybrid model based on discrete wavelet transform (DWT) and bidirectional recurrent neural networks for wind speed prediction,” Engineering Applications of Artificial Intelligence , vol. 127, p. 107340, 2024

2024
[19]

LSTM-autoencoder-based anomaly detection for indoor air quality time-series data,

Y . Wei, J. Jang-Jaccard, W. Xu, F. Sabrina, S. Camtepe, and M. Boulic, “LSTM-autoencoder-based anomaly detection for indoor air quality time-series data,” IEEE Sensors Journal, vol. 23, no. 4, pp. 3787–3800, 2023

2023
[20]

User’s guide for the commercial modular aero-propulsion system simulation (c-mapss),

D. K. Frederick, J. A. DeCastro, and J. S. Litt, “User’s guide for the commercial modular aero-propulsion system simulation (c-mapss),” Tech. Rep., 2007

2007

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Anomaly detection on time series,

M. Teng, “Anomaly detection on time series,” in proceeding of IEEE International Conference on Progress in Informatics and Computing , vol. 1, 2010, pp. 603–608

work page 2010

[2] [2]

Deep learning for anomaly detection in time-series data: Review, analysis, and guidelines,

K. Choi, J. Yi, C. Park, and S. Yoon, “Deep learning for anomaly detection in time-series data: Review, analysis, and guidelines,” IEEE Access, vol. 9, pp. 120 043–120 065, 2021

2021

[3] [3]

Outlier detection and missing value in seasonal ARIMA model using rainfall data,

P. Arumugam and R. Saranya, “Outlier detection and missing value in seasonal ARIMA model using rainfall data,” Materials Today: Proceed- ings, vol. 5, no. 1, pp. 1791–1799, 2018

work page 2018

[4] [4]

Modeling data containing outliers using ARIMA additive outlier (ARIMA-AO),

A. S. Ahmar, S. Guritno, A. Rahman, I. Minggi, M. A. Tiro, M. K. Aidid, S. Annas, D. U. Sutiksno, D. S. Ahmar, K. H. Ahmar et al., “Modeling data containing outliers using ARIMA additive outlier (ARIMA-AO),” in Journal of Physics: Conference Series, vol. 954, no. 1. IOP Publishing, 2018, p. 012010

work page 2018

[5] [5]

A machine learning framework for network anomaly detection using SVM and GA,

T. Shon, Y . Kim, C. Lee, and J. Moon, “A machine learning framework for network anomaly detection using SVM and GA,” in Proceedings from the 6th annual IEEE SMC information assurance workshop. IEEE, 2005, pp. 176–183

work page 2005

[6] [6]

Anomaly detection based on machine learning: dimensionality reduction using PCA and classification using SVM,

A. George and A. Vidyapeetham, “Anomaly detection based on machine learning: dimensionality reduction using PCA and classification using SVM,” International Journal of Computer Applications , vol. 47, no. 21, pp. 5–8, 2012

work page 2012

[7] [7]

A novel anomaly detection method based on adaptive mahalanobis-squared distance and one-class kNN rule for structural health monitoring under environmental effects,

H. Sarmadi and A. Karamodin, “A novel anomaly detection method based on adaptive mahalanobis-squared distance and one-class kNN rule for structural health monitoring under environmental effects,” Mechan- ical Systems and Signal Processing , vol. 140, p. 106495, 2020

work page 2020

[8] [8]

Unsupervised anomaly detection in time series using LSTM-based autoencoders,

O. I. Provotar, Y . M. Linder, and M. M. Veres, “Unsupervised anomaly detection in time series using LSTM-based autoencoders,” in 2019 IEEE International Conference on Advanced Trends in Information Theory (ATIT), 2019, pp. 513–517

work page 2019