Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Enhanced anomaly detection in well log data through the application of ensemble GANs

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that an ensemble of generative adversarial networks trained on univariate well-log measurements flags anomalous readings with higher precision and F1 than a Gaussian mixture model across GR, DT, NPHI, and RHOB logs.

desk verdict A modest, incompletely written comparison: the EGAN method section is placeholder text, so the central claim cannot be evaluated as the paper stands. read the letter →

arxiv 2411.19875 v1 pith:J6PLYECU submitted 2024-11-29 physics.geo-ph cs.AIcs.LG

Abdulrahman Al-Fakih , A. Koeshidayatullah , Tapan Mukerji , SanLinn I. Kaka This is my paper · ORCID
classification physics.geo-phcs.AIcs.LG
keywords WelllogdataEnsembleGANsAnomalydetectionGaussianmixturemodelIsolationforestGenerativeadversarialnetworksReservoirmanagementUnivariateclassification
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

Ensemble generative adversarial networks (EGANs), which combine multiple GANs to learn a data distribution, can flag anomalous well-log readings more precisely than a Gaussian mixture model (GMM). On four univariate logs from two North Sea wells, the paper reports higher precision and F1 for EGANs in every case: gamma ray, sonic travel time, neutron porosity, and bulk density. The GMM tends to overflag, producing more false positives, while the EGAN keeps recall high and reduces those false alarms. The paper positions this as a first application of ensemble GANs to structured well-log tabular data, with practical consequences for screening outliers before drilling and reservoir-management decisions.

What carries the argument

The mechanism is the ensemble GAN, a combination of multiple generative adversarial networks whose discriminator acts as a judge of data authenticity: after training on standardized well-log values, the discriminator assigns low scores to points that deviate from the learned distribution, and a threshold converts those scores into anomaly labels. The GMM baseline fits two full-covariance Gaussian components and flags points in low-probability regions. An isolation forest, applied after K-means filtering to two selected clusters, supplies the reference labels used for precision, recall, and F1.

What would settle it

Take the same two North Sea wells and score EGAN and GMM predictions against independently established geological markers such as core descriptions, formation tops, or drilling incidents; if the GMM matches those markers as often as the EGAN does, the reported precision advantage would be an artifact of the chosen reference labels rather than better detection.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that an ensemble GAN's discriminator learns the distribution of each well-log variable closely enough that readings outside the high-density regions can be separated from normal readings, and that this separation agrees with an Isolation Forest reference better than a two-component GMM does. Reported precision and F1 are 0.62 and 0.76 for gamma ray, 0.70 and 0.79 for sonic travel time, 0.53 and 0.68 for neutron porosity, and 0.52 and 0.67 for bulk density, against 0.38/0.54, 0.56/0.71, 0.47/0.61, and 0.50/0.65 for the GMM. Classification is univariate: the bivariate contour plots illustrate the learned distributions but do not participate in the anomaly labeling.

Load-bearing premise

The load-bearing premise is that the labels produced by the isolation-forest algorithm after K-means filtering are a trustworthy reference for what counts as an anomaly, because every precision and F1 comparison is measured against those labels.

Editorial extensions

If this is right

  • In all four logs tested, EGANs reduced false positives relative to GMM while keeping recall near or above 0.87, so fewer normal readings would be flagged for investigation.
  • The workflow needs no geologically pre-labeled anomalies; the reference labels come from an unsupervised isolation-forest detector after clustering.
  • Because each log is treated independently, the same EGAN pipeline can be applied to any single well-log curve without retraining on multivariate data.
  • The EGAN's confidence threshold can be adjusted, letting an operator make the detector more or less sensitive for a given drilling or reservoir-management context.
  • The results imply that the practical bottleneck for well-log anomaly screening is false positives, and that a distribution-learning model can address it better than a parametric mixture model.

Reading between the lines

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

  • The paper does not validate its anomaly flags against independent geological markers, so the stated precision advantage is a measure of agreement with one unsupervised reference detector rather than proven field accuracy.
  • Section 2.4.2 contains placeholder text where the GAN architecture should be described, so the reported numbers can be reproduced only from the linked code repository, not from the manuscript text alone.
  • The multivariate structure visible in the contour plots is deliberately excluded from classification; a single EGAN fed all four logs together could plausibly catch anomalies that no one curve exposes.
  • If the same discriminator-scoring idea transfers, it could rank anomalies in other depth-indexed measurements such as deep resistivity, where labeled geological anomalies are scarce.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes using ensemble generative adversarial networks (EGANs) to detect anomalies in univariate well log data (GR, DT, NPHI, RHOB) from two North Sea wells, and compares performance against Gaussian mixture models (GMMs). The pipeline first applies K-means clustering, retains two clusters, then uses an Isolation Forest (IF) algorithm to label points as normal or anomalous. GMM and EGAN models are trained on these labels, and precision, recall, and F1 scores are reported in Table 4, claiming EGANs generally achieve higher precision and F1. The central claim is that EGANs outperform GMMs for this anomaly detection task.

Significance. If the claim were well-supported, the paper would provide a useful comparison of modern generative anomaly detection against a classical baseline in a geophysical application, a topic with practical relevance for reservoir management. Credit is due for releasing code and data on GitHub, which enables inspection. However, the evaluation is anchored to labels produced by an unsupervised algorithm (Isolation Forest) rather than independent ground truth, and the EGAN method itself is not described in the manuscript. As submitted, the paper does not support its headline claim because the method is undefined and the reported metrics measure agreement with one particular detector. The falsifiable prediction that EGANs beat GMMs on these four logs is interesting, but this version does not allow the reader to verify or reproduce it from the text.

major comments (3)
  1. [Section 2.4.2] The technical description of the EGAN model is absent: after the heading 'Generative adversarial networks model', the text contains lorem ipsum placeholder paragraphs ('Suspendisse vel felis...') followed by generic GAN background and equations for the original GAN. No ensemble aggregation strategy is described, no anomaly scoring rule is given, and no procedure is provided for combining multiple GANs, despite Table 2 listing hyperparameters for only a single generator/discriminator pair. Because the central claim in Table 4 concerns the performance of EGANs, the method that produced those numbers is undefined in the manuscript. This is not a minor omission; the core contribution cannot be evaluated.
  2. [Sections 2.1 and 3.3] The evaluation labels are generated by the Isolation Forest algorithm on the same data that is later used for training and testing the GMM and EGAN models. Table 4 therefore reports how well each model agrees with IF's anomaly assignments, not how well they detect true geological anomalies. The paper provides no independent validation of the IF labels (e.g., via core data, production tests, or known stratigraphic markers). Consequently, the claimed superiority of EGANs, such as the GR precision of 0.62 versus 0.38 in Table 4, is a statement of concordance with one particular algorithmic definition of 'anomaly', and the circularity weakens the central conclusion.
  3. [Sections 2.5, 3.1, and 3.2] The classification threshold for converting model scores into anomaly labels is never specified. The text repeatedly refers to 'a defined probability threshold' or points 'above or below a defined probability threshold', but no threshold values, selection criterion, or sensitivity analysis are given for either GMM or EGAN. Since precision, recall, and F1 in Table 4 depend directly on these thresholds, the reported scores cannot be reproduced or interpreted from the manuscript, and the GMM-vs-EGAN comparison may hinge entirely on an undocumented threshold choice.
minor comments (6)
  1. [Title and Introduction] The title contains an extraneous space in 'Ensemble GAN s'; the authors should correct this typographical error.
  2. [Introduction, paragraph 3] The sentence 'ensemble generative adversarial networks (EGANs) was introduced' cites reference [14], which is the original Goodfellow GAN paper, not a source for ensemble GANs. The authors should cite the GAN ensemble literature, such as reference [15], and clarify which specific ensemble construction is used.
  3. [Table 2] The hyperparameter table for EGANs lists learning rates, batch size, and layer sizes for a single GAN, but omits the number of GANs in the ensemble and any ensemble-specific parameters (e.g., voting scheme, weighting). This makes the table inconsistent with the 'ensemble' claim.
  4. [Section 5] The conclusion states that the work demonstrates 'anomaly detection outside the data distribution and forecasting in geophysics', but no forecasting experiment is presented; the word 'forecasting' should be removed.
  5. [Abbreviations] The abbreviations list includes 'NixtlaClient', which does not appear anywhere in the paper; this appears to be a leftover from another project and should be deleted.
  6. [References] Reference [1] is titled 'Application of GAN for Reducing Data Imbalance under Limited Dataset' but the journal-style citation expands 'Gamma Ray Graphics Theory and Applications' in a way that appears to insert an unrelated field name; the authors should verify the accuracy of this citation.

Circularity Check

1 steps flagged · score 6.0 of 10

Partial circularity: Table 4's anomaly-detection comparison is scored against Isolation Forest pseudo-labels generated on the same K-Means-filtered data, so 'outperforming' reduces to matching IF; Section 2.4.2 is placeholder text and leaves the EGAN method undefined.

  1. self definitional [Section 2.1 (IF labeling) and Section 3.3 / Table 4 (evaluation against labeled data)]
    "The isolation forest (IF) algorithm was employed after the clustering process to detect anomalies within the filtered data (from clusters 0 and 1). The IF algorithm flags potential anomalies by labeling data points as either -1 (anomalies) or 1 (normal). ... The performance of EGANs and GMM was evaluated across several datasets using key metrics—precision, recall, and F1 score."

    The labels used to compute precision, recall, and F1 (Eqs. 4-6) are not independent geological truth; they are the output of an Isolation Forest applied to the same K-Means-filtered well-log data. The evaluation target is therefore by construction the IF anomaly assignment. Claiming EGANs 'outperform' GMM in anomaly detection reduces to claiming they agree more with IF's definition of anomalies on the same data, not that they better detect true subsurface anomalies. The benchmark is internally defined by the detector being used as reference, so the Table 4 comparison is a detector-consistency check rather than an externally grounded prediction.

full rationale

The main circular step is the pseudo-ground-truth construction. Section 2.1 defines anomalies via Isolation Forest labeling of the same KMeans-filtered data; Section 3.3 then evaluates both models with precision/recall/F1 against those labels. Thus the claimed superiority is a measure of agreement with IF, not with independently known anomalies. This is a partial, structural circularity because the scores are not numerically forced, but the evaluation target is defined by an algorithm on the same data. Separately, Section 2.4.2 contains placeholder text ('Suspendisse vel felis...') followed by a generic GAN description; no ensemble aggregation rule, anomaly scoring function, or threshold-selection procedure is given. This is an omitted method / missing support rather than a circularity per se, but it means the EGAN column of Table 4 has no defined derivation in the manuscript and cannot be independently checked. No load-bearing self-citation chain was found; the references to the authors' prior GAN/GMM work are contextual, not the basis of the comparison.

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

The central claim rests on several ad hoc choices: the number of K-Means clusters, which clusters are retained, the anomaly thresholds, and most importantly the use of Isolation Forest outputs as ground truth. The EGAN architecture itself is not described due to placeholder text, so the model configuration is an unverified assumption. No new physical entities are introduced.

free parameters (6)
  • Number of K-Means clusters = 10
    Selected by visual inspection of patterns and metric consistency; not optimized or justified quantitatively.
  • K-Means clusters retained = 0 and 1
    Clusters 0 and 1 were chosen as 'relevant' and all other clusters discarded; this post hoc filtering shapes the anomaly distribution.
  • GMM number of components = 2
    Set to 2 for all logs; no model selection criterion reported.
  • Anomaly probability threshold for GMM/EGAN = not specified
    The paper refers to a 'defined probability threshold' but never gives its value; reported precision/recall depend directly on it.
  • EGAN learning rates = 0.001 - 0.002
    Given as ranges; no tuning procedure or final values reported.
  • EGAN batch size = 32 - 64
    Given as a range; no specific value reported.
assumptions (4)
  • domain assumption K-Means clustering with k=10 produces meaningful clusters for the well log data
    The choice of 10 clusters and retention of clusters 0 and 1 is based on visual assessment, not a formal criterion. It determines which data points enter the anomaly detection task.
  • ad hoc to paper The Isolation Forest algorithm generates valid ground-truth anomaly labels
    The labels from IF are used as targets for evaluating GMM and EGAN, but no independent validation (e.g., known geological markers) is provided. This is the load-bearing premise for all reported scores.
  • domain assumption Standard GAN training converges for tabular well log data with the listed hyperparameters
    The GAN architecture and training dynamics are not described in Section 2.4.2 (placeholder text), so the reader must assume the model trains properly.
  • standard math Gaussian mixture model assumption
    The GMM assumes normality of the data distribution, a standard statistical assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhanced anomaly detection in well log data through the application of ensemble GANs." pith.science (2026). https://pith.science/paper/J6PLYECU

@misc{pith2026241119875,
  author       = {Pith},
  title        = {Pith review of: Enhanced anomaly detection in well log data through the application of ensemble GANs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J6PLYECU}},
  note         = {Machine review of arXiv:2411.19875}
}
read the original abstract

Although generative adversarial networks (GANs) have shown significant success in modeling data distributions for image datasets, their application to structured or tabular data, such as well logs, remains relatively underexplored. This study extends the ensemble GANs (EGANs) framework to capture the distribution of well log data and detect anomalies that fall outside of these distributions. The proposed approach compares the performance of traditional methods, such as Gaussian mixture models (GMMs), with EGANs in detecting anomalies outside the expected data distributions. For the gamma ray (GR) dataset, EGANs achieved a precision of 0.62 and F1 score of 0.76, outperforming GMM's precision of 0.38 and F1 score of 0.54. Similarly, for travel time (DT), EGANs achieved a precision of 0.70 and F1 score of 0.79, surpassing GMM 0.56 and 0.71. In the neutron porosity (NPHI) dataset, EGANs recorded a precision of 0.53 and F1 score of 0.68, outshining GMM 0.47 and 0.61. For the bulk density (RHOB) dataset, EGANs achieved a precision of 0.52 and an F1 score of 0.67, slightly outperforming GMM, which yielded a precision of 0.50 and an F1 score of 0.65. This work's novelty lies in applying EGANs for well log data analysis, showcasing their ability to learn data patterns and identify anomalies that deviate from them. This approach offers more reliable anomaly detection compared to traditional methods like GMM. The findings highlight the potential of EGANs in enhancing anomaly detection for well log data, delivering significant implications for optimizing drilling strategies and reservoir management through more accurate, data-driven insights into subsurface characterization.

Figures

Figures reproduced from arXiv: 2411.19875 by the authors.

Figure 1
Figure 1. Visualization of dataset distributions and clustering results for (a) GR, (b) DT, (c) RHOB, and (d) NPHI logs. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Workflow for anomaly detection in well log data using (a) Gaussian Mixture Model (GMM) and (b) Ensemble [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. GMM analysis of well log data (GR, DT, RHOB, NPHI). The leftmost columns display scatter plots of the [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: EGAN analysis of well log data (GR, RHOB, DT, NPHI). The first column shows scatter plots of the datasets [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Leveraging Time-Series Foundation Model for Subsurface Well Logs Prediction and Anomaly Detection

    physics.geo-ph 2024-12 reject novelty 4.0 of 10

    Fine-tuned TimeGPT forecasts some well logs reasonably, but the claimed 93% anomaly detection accuracy is contradicted by ROC-AUC near 0.51 and MCC near zero.

Reference graph

Works this paper leans on

38 extracted references · 24 canonical work pages · cited by 1 Pith paper

  1. [1]

    G. Adke. Application of GAN for Reducing Data Imbalance under Limited Dataset. In Proceedings of the International Joint Conference on Computer Vision, Imaging, and Computer Gamma Ray Graphics Theory and Applications, pages 4, 2022. https://doi.org/10.5220/0010782800003124

    work page doi:10.5220/0010782800003124 2022

  2. [2]

    Al-Fakih, S

    A. Al-Fakih, S. I. Kaka, and A. I. Koeshidayatullah. Reservoir Property Prediction in the North Sea Using Machine Learning. IEEE Access, 11, 2023. https://doi.org/10.1109/ACCESS.2023.3336623

    arXiv 2023

  3. [3]

    Al-Fakih, A

    A. Al-Fakih, A. Koeshidayatullah, and S. Kaka. AI-Driven Reservoir Management: GANs and GMM for Enhanced Control. In ECMOR 2024, volume 2024, number 1, pages 1–11, 2024. https://doi.org/10.3997/2214- 4609.202437004

    arXiv 2024

  4. [4]

    Al-Fakih, S

    A. Al-Fakih, S. I. Kaka, and A. Koeshidayatullah. Utilizing GANs for Synthetic Well Logging Data Generation: A Step Towards Revolutionizing Near-Field Exploration. In EAGE/AAPG Workshop on New Discoveries in Mature Basins, volume 2024, number 1, pages 1–5, 2024. https://doi.org/10.3997/2214-4609.202471016

    work page arXiv 2024

  5. [5]

    Al-Fakih, A

    A. Al-Fakih, A. Koeshidayatullah, and S. Kaka. Enhancing Geoscience Analysis: AI-Driven Imputation of Missing Data in Well Logging Using Generative Models. In EGU General Assembly 2024, Vienna, Austria, 14–19 April 2024, EGU24-10627, 2024. https://doi.org/10.5194/egusphere-egu24-10627

    work page doi:10.5194/egusphere-egu24-10627 2024

  6. [6]

    Minimax Theory for High-dimensional Gaussian Mixtures with Sparse Mean Separation

    M. Azizyan, A. Singh, and L. A. Wasserman. Minimax Theory for High-dimensional Gaussian Mixtures with Sparse Mean Separation. In Neural Information Processing Systems, 2013. https://doi.org/10.48550/arXiv.1306.2035

    work page Pith review arXiv doi:10.48550/arxiv.1306.2035 2013

  7. [7]

    Kushwaha and G

    V . Kushwaha and G. C. Nandi. Study of Prevention of Mode Collapse in Generative Adversarial Net- work (GAN). In 4th IEEE Conference on Information and Communication Technology, CICT 2020 , 2020. https://doi.org/10.1109/CICT51604.2020.9312049

    arXiv 2020

  8. [8]

    Bourou, A

    S. Bourou, A. El Saer, T. H. Velivassaki, A. V oulkidis, and T. Zahariadis. A Review of Tabular Data Synthesis Using GANs on an IDS Dataset. Information (Switzerland), 12(9), 2021. https://doi.org/10.3390/info12090375

    doi:10.3390/info12090375 2021

Show all 38 references
  1. [9]

    Stauffer and W

    C. Stauffer and W. E. L. Grimson. Adaptive Background Mixture Models for Real-Time Tracking. In Proceedings. 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149), volume 2, pages 246–252, Fort Collins, CO, USA, 1999. doi: 10.1109/C...

    1999

  2. [10]

    T. Darling. Well Logging and Formation Evaluation. In Well Logging and Formation Evaluation , 2005. https://doi.org/10.1016/B978-0-7506-7883-4.X5000-1

    2005 doi

  3. [11]

    L. A. B. de Oliveira and C. de C. Carneiro. Synthetic Geochemical Well Logs Generation Using Ensemble Machine Learning Techniques for the Brazilian Pre-Salt Reservoirs. Journal of Petroleum Science and Engineering, 196:108080, 2021. https://doi.org/10.1016/j.petrol.2020.108080

    2021

  4. [12]

    Fernandes, K

    W. Fernandes, K. S. Komati, and K. Assis de Souza Gazolli. Anomaly Detection in Oil-Producing Wells: A Comparative Study of One-Class Classifiers in a Multivariate Time Series Dataset.Journal of Petroleum Exploration and Production Technology, 14(1), 2024. https://doi.org/10.1...

    2024 doi

  5. [13]

    Y . Gan, J. Cao, Y . Lu, Y . He, and H. Wang. Reservoir Prediction Based on Stacked Denoising Auto-Encoder for Feature Extraction. In SEG Global Meeting Abstracts, pages 1736–1739, 2018. https://doi.org/10.1190/IGC2018- 426

    2018 doi

  6. [14]

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio. Generative Adversarial Nets. InProceedings of the 27th International Conference on Neural Information Processing Systems (NIPS’14), pages 2672–2680, MIT Press, 2014. h...

    2014

  7. [15]

    X. Han, X. Chen, and L.-P. Liu. GAN Ensemble for Anomaly Detection. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, number 5, pages 4090–4097, 2021. https://doi.org/10.1609/aaai.v35i5.16530. 13 A PREPRINT - DECEMBER 2, 2024

    2021 doi

  8. [16]

    M. Ibrahim. Precision vs. Recall: Understanding How to Classify with Clarity. Retrieved from https://wandb.ai/mostafaibrahim17/ml-articles/reports/Precision-vs-Recall-Understanding-How-to-Classify- with-Clarity–Vmlldzo1MTk1MDY5, 2021

    2021

  9. [17]

    Gokcesu, M

    K. Gokcesu, M. M. Neyshabouri, H. Gokcesu, and S. S. Kozat. Sequential Outlier Detection Based on Incremental Decision Trees. IEEE Transactions on Signal Processing, 67(4), 993–1005, 2019. doi: 10.1109/TSP.2018.2887406

    2019

  10. [18]

    Kazemi and H

    A. Kazemi and H. Meidani. IGANI: Iterative Generative Adversarial Networks for Imputation with Application to Traffic Data. IEEE Access, 9, 2021. https://doi.org/10.1109/ACCESS.2021.3103456

    2021

  11. [19]

    K. Klu. F-Score: What are Accuracy, Precision, Recall, and F1 Score? Retrieved from https://klu.ai/f-score-what- are-accuracy-precision-recall-and-f1-score/, 2023

    2023

  12. [20]

    J. Lai, Y . Su, L. Xiao, F. Zhao, T. Bai, Y . Li, H. Li, Y . Huang, G. Wang, and Z. Qin. Application of Geophysical Well Logs in Solving Geologic Issues: Past, Present and Future Prospect. Geoscience Frontiers, 15(3), 2024. https://doi.org/10.1016/j.gsf.2024.101779

    2024

  13. [21]

    Landauer, S

    M. Landauer, S. Onder, F. Skopik, and M. Wurzenberger. Deep Learning for Anomaly Detection in Log Data: A Survey. Machine Learning with Applications, 12, 100470, 2023. https://doi.org/10.1016/j.mlwa.2023.100470

    2023

  14. [22]

    W. Lim, K. S. C. Yong, B. T. Lau, and C. C. L. Tan. Future of Generative Adversarial Networks (GAN) for Anomaly Detection in Network Security: A Review. Computers & Security , 139, 103733, 2024. https://doi.org/10.1016/j.cose.2024.103733

    2024

  15. [23]

    S. M. Luthi. Geological Well Logs. In Geological Well Logs , Springer Berlin Heidelberg, 2001. https://doi.org/10.1007/978-3-662-04627-2

    2001 doi

  16. [24]

    J. Liu, H. Zhu, Y . Liu, H. Wu, Y . Lan, and X. Zhang. Anomaly Detection for Time Series Using Temporal Convolutional Networks and Gaussian Mixture Model. In Journal of Physics: Conference Series, volume 1187, number 4, page 042111, IOP Publishing, 2019. https://doi.org/10.108...

    2019 doi

  17. [25]

    Y . Z. Ma. Petrophysical Data Analytics for Reservoir Characterization. In Quantitative Geosciences: Data Analytics, Geostatistics, Reservoir Characterization and Modeling, 2019. https://doi.org/10.1007/978-3-030-17860- 49

    2019 doi

  18. [26]

    Martí, N

    L. Martí, N. Sanchez-Pi, J. M. Molina, and A. C. B. Garcia. Anomaly Detection Based on Sensor Data in Petroleum Industry Applications. Sensors, 15(2):2774–2797, 2015. https://doi.org/10.3390/s150202774

    2015 doi

  19. [27]

    Mishra, A

    A. Mishra, A. Sharma, and A. K. Patidar. Evaluation and Development of a Predictive Model for Geophysical Well Log Data Analysis and Reservoir Characterization: Machine Learning Applications to Lithology Prediction. Natural Resources Research, 31(6), 2022. https://doi.org/10.1...

    2022 doi

  20. [28]

    Peer and J

    U. Peer and J. G. Dy. Automated Target Detection for Geophysical Applications.IEEE Transactions on Geoscience and Remote Sensing, 55(3), 2017. https://doi.org/10.1109/TGRS.2016.2627245

    2017

  21. [29]

    D. M. W. Powers. Evaluation: From Precision, Recall and F-Factor to ROC, In-formedness, Markedness & Correlation. Tech. Rep. SIE-07-001, School of Informatics and Engineering, Flinders University, Adelaide, Australia, 2007. https://doi.org/10.48550/arXiv.2010.16061

    2007 doi arXiv

  22. [30]

    Rebala, A

    G. Rebala, A. Ravi, and S. Churiwala. An Introduction to Machine Learning. Springer, 2019. https://doi.org/10.1007/978-3-030-15729-6

    2019 doi

  23. [31]

    Reynolds

    D. Reynolds. Gaussian Mixture Models. In Encyclopedia of Biometrics , pages 196–202. Springer, 2015. https://doi.org/10.1007/978-1-4899-7488-4-196

    2015 doi

  24. [32]

    Improving Upon Precision, Recall, and F1 with Gain Metrics

    Snorkel AI. Improving Upon Precision, Recall, and F1 with Gain Metrics. Retrieved from https://snorkel.ai/precision-recall-f1-gain-metrics/, 2022

    2022

  25. [33]

    Struminskiy, A

    K. Struminskiy, A. Klenitskiy, A. Reshytko, D. Egorov, A. Shchepetnov, A. Sabirov, D. Vetrov, A. Semenikhin, O. Osmonalieva, and B. Belozerov. Well Log Data Standardization, Imputation and Anomaly Detection Using Hidden Markov Models. In 4th EAGE Conference on Petroleum Geosta...

    2019

  26. [34]

    M. B. Valentín, C. R. Bom, A. L. M. Compan, M. D. Correia, C. M. de Jesus, A. L. de Souza, M. P. de Albuquerque, and E. L. Faria. Estimation of Permeability and Effective Porosity Logs Using Deep Autoencoders in Borehole Image Logs from the Brazilian Pre-Salt Carbonate. Journa...

  27. [35]

    C. Zhao, J. Zhao, W. Wang, C. Yuan, and J. Tang. A Novel Hybrid Ensemble Model for Mineral Prospectivity Prediction: A Case Study in the Malipo W-Sn Mineral District, Yunnan Province, China. Ore Geology Reviews, 168:106001, 2024. https://doi.org/10.1016/j.oregeorev.2024.106001...

    2024

  28. [36]

    Zhang, R

    Y . Zhang, R. Zhang, and B. Zhao. A Systematic Review of Generative Adversarial Imputation Network in Missing Data Imputation. Neural Computing and Applications, 35(27), 2023. https://doi.org/10.1007/s00521-023-08840-2

    2023 doi

  29. [37]

    B. Zong, Q. Song, M. R. Min, W. Cheng, C. Lumezanu, D. Cho, and H. Chen. Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection. In Proceedings of the International Conference on Learning Representations, 2018. https://openreview.net/forum?id=SJUdkecgx. 15

    2018

  30. [2018]

    https://doi.org/10.1016/j.petrol.2018.06.038

    2018 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.