Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Multiple data-driven missing imputation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a position-aware, rule-based imputer for short gaps in univariate time series outperforms standard imputation methods by roughly 37 percent average reconstruction error.

desk verdict A transparent rule-based imputer with a checkable implementation, but the paper's headline superiority claim doesn't survive its own Table 1. read the letter →

arxiv 2507.03061 v1 pith:E4GCDT55 submitted 2025-07-03 stat.ME cs.LGstat.ML

Sergii Kavun This is my paper
classification stat.MEcs.LGstat.ML MSC 62M1062D10
keywords missingdataimputationtimeseriesKZImputergapanalysispreprocessingunivariateboundarygapshighsparsity
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 claims that where a missing block sits in a univariate time series, and how long it is, should determine how it is filled, and that simple local means built on that idea can beat standard imputation baselines. The proposed method, KZImputer, handles gaps of one to five points with separate formulas for the start, middle, and end of the series, using averaged values from nearby observed points and, for larger gaps, previously imputed cells. The reported experiments put the average reconstruction-error improvement over the tested baselines at roughly 37.3 percent, with the largest gains for single-point boundary gaps and smaller gains for five-point gaps. If the claim holds, KZImputer gives analysts a transparent, fast, model-free way to prepare short-gap series for forecasting and anomaly detection, especially when missingness is high.

What carries the argument

The carrying object is a family of position- and length-specific averaging formulas built on a cached_mean helper that averages only available non-NaN points. A single middle gap is filled by the mean of its two neighbours; start and end gaps use the mean of the next three or previous three observed points respectively. Double and triple gaps are filled from the boundary inward, and quadruple and pentuple gaps impute the outer cells first, then form inner cells as means of already imputed values. That stepwise recursion—treating an earlier imputation as an observed value when computing a later cell—is what lets the method stretch local information across a short block, and it is also the mechanism that carries the method's main risk.

What would settle it

Run KZImputer against linear interpolation and forward/backward fill on series with known ground truth, separating middle gaps from boundary gaps, and compare per-gap errors to Table 1; a single series with a sharp trend or seasonal peak at the boundary should expose whether the three- or five-point means mislocate the missing value, because those means are pulled toward the observed block and away from the true local level.

Watch

Extended reading notes

Core claim

KZImputer's central claim is that explicit positional adaptation—distinct averaging formulas for left, middle, and right gaps of lengths one through five—lets a small set of local arithmetic means outperform established imputation methods (mean, median, forward/backward fill, linear and spline interpolation, k-nearest neighbors, and iterative imputation) on short-gap reconstruction. The paper reports an average imputation-error improvement of about 37.3 percent over the tested baselines, with single-point gaps improving by roughly 48 percent over backward fill and 41 percent over forward fill, and with a smaller but positive edge retained for five-point gaps. The method is aimed at univariate series in high-sparsity regimes, around 50 percent missingness or more, where simple baselines degrade, and it is not claimed to handle long gaps or multivariate dependencies.

Load-bearing premise

The load-bearing premise is that, when filling a gap cell by cell, a value the method just imputed can be treated as if it were really observed; if an early imputation is off, the error feeds into every later cell in the same gap, and the paper does not analyze how much that feedback inflates or masks the reported gains.

Editorial extensions

If this is right

  • Short boundary gaps of one to five points can be reconstructed more accurately by directional local means than by global constants or simple forward/backward fill.
  • The method's advantage is clearest in high-sparsity settings, where conventional baselines degrade, and it stays cheap enough for resource-constrained or online preprocessing.
  • Post-imputation series preserve distributional and correlational structure better than the tested baselines, according to the reported divergence and correlation-difference metrics.
  • The roughly 37 percent average error improvement makes KZImputer a plausible default preprocessing step for univariate sensor, energy, and financial series before forecasting, provided gaps stay short.

Reading between the lines

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

  • If the empirical claim transfers, the same position-aware design could be extended to adapt the averaging window to local trend or seasonality rather than fixed three- and five-point arms, which might preserve the gains on non-stationary series.
  • The unanalyzed feedback in the stepwise formulas invites a simple test: compare the one-pass version against an iterated version that re-imputes the block until it stabilizes, to see whether treating earlier imputations as observed helps or hurts.
  • Because the method is transparent and parameter-light, it could serve as a sanity-check baseline for deep imputation models on short gaps, clarifying the regimes where extra model complexity actually buys accuracy.
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

4 major / 5 minor

Summary. The manuscript proposes KZImputer, a rule-based imputation method for univariate time series with consecutive missing gaps of length 1 to 5, with distinct formulas for gaps at the left, middle, and right of the series. The method computes local averages of neighboring observed values, using stepwise replacement within a gap, and the paper claims that it consistently outperforms standard imputation methods, with a mean improvement of about 37.3%, particularly at high missingness rates, based on Table 1 and several figures. The paper also discusses comparison methods, evaluation metrics, and application scenarios such as wind energy and air quality data.

Significance. KZImputer is transparent, computationally light, and the authors provide code through a repository (Ref. 49), which are genuine strengths for a method aimed at short boundary gaps. If the empirical claims were substantiated, a simple non-iterative imputer tailored to short gaps at the start or end of a series could be a useful practical baseline. However, the significance is not currently established: the reported evidence is internally inconsistent, the experimental description is written in the future tense rather than as a completed study, and the central claim of consistent superiority is contradicted by the paper's own Table 1. The availability of pseudocode and a repository is a positive feature, but the manuscript as submitted does not provide a reliable basis for the stated conclusions.

major comments (4)
  1. [Section 5, Table 1] Table 1 contradicts the central claim that KZImputer 'consistently outperforms standard imputation methods.' For LENGAP=4, k=5, the proposed error is 51.85% versus 44.29% for backward fill and 45.40% for forward fill, giving improvements of -17.05% and -14.19%; for LENGAP=2, k=2, the proposed error is 20.33% versus 19.58% for forward fill, giving -3.79%. Averaging the six backward improvements gives about 22.6% and the six forward improvements about 23.0%; no aggregation of the table's numbers yields the claimed 'approximately 37.3%' mean improvement. The paper's headline empirical result is therefore not supported by its own data.
  2. [Section 4 (Experimental environment), Section 4.4, Eq. (4)] The experimental section is written in the future tense ('is intended for use', 'would be introduced', 'will be benchmarked', 'will be assessed'), which makes it unclear whether the reported results were actually obtained. Section 4.4 states that details of the quadruple-gap middle case are 'omitted for brevity', and Eq. (4) contains malformed notation: sums with lower bounds exceeding upper bounds, undefined piecewise conditions such as 'else odd integers', and references to SET_rev L that are not defined. These omissions prevent an independent implementation and regeneration of Table 1 and Figures 5-14.
  3. [Section 4.2 and 4.3, Eqs. (2)-(3)] The stepwise formulas treat previously imputed values as observed when computing later cells. For example, in Eq. (2) the left-end double-gap formula for GAP_{i=1}^L averages observed VAR values together with GAP_{i=LENGAP}^L, and Eq. (3) for the triple-gap left-end case includes sums over GAP_j^L. If the first imputed value is biased, the error propagates to subsequent cells, yet the paper provides no bias, variance, or error-propagation analysis. The claimed superiority over baselines relies on this recursive feedback being harmless in the tested configurations, which is not demonstrated.
  4. [Section 5, Figs. 6-14] The results are reported as point values without confidence intervals, standard deviations, or significance tests. The text claims '100 experiments (cross-validation approach) with 100 randomly corrupted data positions', but Table 1 reports only single numbers per configuration, and Figures 6-14 appear to show a single dataset each. Without variability or significance information, negative improvements such as those in Table 1 cannot be dismissed as noise, and the superiority claim is unverifiable.
minor comments (5)
  1. [Section numbering] The methodology section (Section 3) contains subsections numbered 4.1-4.5, while the following 'Experimental environment' section also restarts at 4.1; this impairs navigation and suggests a structural error in the manuscript.
  2. [Literature Review] The Literature Review contains placeholder texts 'see the placeholder for a reference on the limitations of mean imputation' and 'see the placeholder for a reference on regression imputation', which should be replaced with actual citations.
  3. [References] References 10, 11, and 12 are the same MICE paper, and references 15 and 56 duplicate the BRITS paper; reference [46] also contains an untranslated error message 'Ошибка! Источник ссылки не найден.'
  4. [Section 4.4, three-thirds rule] The 'three thirds' rule is defined for a 26-cell row, and it is unclear how it generalizes to arbitrary series lengths; the rule also appears to overlap with the earlier left/middle/right classification, so its added value and implementation are difficult to determine.
  5. [Eq. (1)] The middle-gap formula in Eq. (1) appears inconsistent with the stated 'mean of the two adjacent cells' rule: with k=3, the displayed sum ranges over j=i-1 to i+k-1, which includes more than two terms; please revise the notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: KZImputer is defined by explicit local-averaging rules and evaluated against held-out ground truth; no prediction or result reduces to its own inputs.

full rationale

The derivation chain is self-contained. The imputation formulas (Eqs. 1-4) are explicit functions of observed neighboring values and, in multi-gap cases, of previously imputed cells; they are not fitted to the evaluation data. The window size k and max_gap_size are stated a priori ('for our research purposes, we set max_gap_size = 5'), with k assigned per LENGAP in the formula blocks, not estimated from the test set. The performance comparison uses original complete series as ground truth under artificially introduced MCAR missingness, so the reported errors are measured against external data rather than constructed by the method's definitions. The author's own references ([28], [29], [49]) are confined to motivation, prior literature, and the code repository; no uniqueness claim or central premise is imported from them. The main weakness of the paper is evidentiary, not circular: Table 1's stated improvements do not aggregate to the claimed 37.3% mean improvement, and two rows show KZImputer worse than forward/backward fill. Those are internal-consistency and correctness problems, not reductions of a result to its inputs. The recursive use of freshly imputed values in Eqs. (2)-(3) may propagate error, but it is a modeling assumption, not a circular definition of the evaluation target.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on hand-set window sizes (k values), an MCAR evaluation design, the assumption that recursively imputed values can be reused as observed, and the assumption that local windows capture the signal. None of these are derived or stress-tested in the paper, and one of them is only justified by a specific 26-cell example.

free parameters (3)
  • max_gap_size = 5 (default)
    Limits the method to gaps of at most five consecutive points; chosen a priori, not from data.
  • shoulder size k for each gap length = k=3 for gap 1, k=4 for gap 2, k=5 for gaps 3-5; Table 1 also lists k=1 for gap 1
    Hand-selected averaging-window lengths in Eqs. (1)-(4); no derivation or tuning procedure, and Section 4.1 conflicts with Table 1 for single gaps.
  • three-thirds boundary rule = Row positions split into thirds (e.g., 1-9, 10-18, 19-26 for a 26-cell row)
    Ad hoc rule in Section 4.4 for inner cells of 4-point middle gaps; based on a specific row length, generalization unclear.
assumptions (4)
  • domain assumption Artificial missingness is Missing Completely At Random (MCAR)
    Section 4.1 says missing data 'would typically be generated under a MCAR mechanism'; real missing data may be MAR or MNAR.
  • ad hoc to paper Previously imputed values can be treated as observed values in later averages
    Eqs. (2)-(4) substitute already imputed GAP values into subsequent averaging formulas; error propagation is not modeled.
  • ad hoc to paper Local windows around the gap contain sufficient information for reconstruction
    The fixed-window averaging rules assume no strong trend, seasonality, or regime shift within the window; Table 2 later admits sensitivity to such structures.
  • ad hoc to paper The 'three thirds' position rule transfers to rows of arbitrary length
    Section 4.4 defines the rule on a 26-cell example and gives no general formula for arbitrary n.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multiple data-driven missing imputation." pith.science (2026). https://pith.science/paper/E4GCDT55

@misc{pith2026250703061,
  author       = {Pith},
  title        = {Pith review of: Multiple data-driven missing imputation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E4GCDT55}},
  note         = {Machine review of arXiv:2507.03061}
}
read the original abstract

This paper introduces KZImputer, a novel adaptive imputation method for univariate time series designed for short to medium-sized missed points (gaps) (1-5 points and beyond) with tailored strategies for segments at the start, middle, or end of the series. KZImputer employs a hybrid strategy to handle various missing data scenarios. Its core mechanism differentiates between gaps at the beginning, middle, or end of the series, applying tailored techniques at each position to optimize imputation accuracy. The method leverages linear interpolation and localized statistical measures, adapting to the characteristics of the surrounding data and the gap size. The performance of KZImputer has been systematically evaluated against established imputation techniques, demonstrating its potential to enhance data quality for subsequent time series analysis. This paper describes the KZImputer methodology in detail and discusses its effectiveness in improving the integrity of time series data. Empirical analysis demonstrates that KZImputer achieves particularly strong performance for datasets with high missingness rates (around 50% or more), maintaining stable and competitive results across statistical and signal-reconstruction metrics. The method proves especially effective in high-sparsity regimes, where traditional approaches typically experience accuracy degradation.

Discussion (0). Sign in 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. Hybrid activation functions for deep neural networks: S3 and S4 -- a novel approach to gradient flow optimization

    cs.LG 2025-07 reject novelty 3.0 of 10

    A new parameterized activation function, S4, that blends sigmoid and softsign through a smooth sigmoid-weighted transition is claimed to improve accuracy and convergence on small neural network benchmarks.

Reference graph

Works this paper leans on

95 extracted references · 66 canonical work pages · cited by 1 Pith paper

  1. [1]

    Missing data are common in real-world datasets, and time series data are especially susceptible

    Introduction Missing data are a pervasive issue in time series analysis, significantly distorting analytical outcomes and compromising forecasting accuracy. Missing data are common in real-world datasets, and time series data are especially susceptible. Missing observations in a time series can arise from various sources, including sensor malfunctions, tr...

  2. [2]

    Literature Review The problem of missing data has been extensively studied. Previous authors have analyzed existing imputation methods [ 4, 44, 20, 45], and a wide array of imputation techniques have been proposed, each with its own set of assumptions, strengths, and weaknesses. Rubin formalized the Missing Data Study [ 46] and introduced multiple imputat...

  3. [3]

    imputation of missing data

    Methodology: KZImputer The KZImputer algorithm provides an adaptive strategy for imputing missing values in univariate time series, specifically targeting consecutive gaps of one to five points (configurable via `max_gap_size`, default = 5). Its core logic differentiates imputation based on the gap's location within the series – referred to as 'left' (sta...

  4. [4]

    This section details the datasets used, the imputation methods chosen for comparison, and the metrics employed for performance assessment

    Experimental environment To evaluate the performance of the KZImputer method, a comprehensive experimental study was designed. This section details the datasets used, the imputation methods chosen for comparison, and the metrics employed for performance assessment. Pseudocode for KZImputer is provided in the Supplementary Materials and the authors’ reposi...

  5. [5]

    a comprehensive evaluation of the results. Our work's practical value lies in its ability to analyze and predict different physical processes in a big data environment, even when dealing with incomplete datasets or data aggregation from multiple sources. Our approach enhances the reliability and accuracy of statistical reports and forecasting models for w...

  6. [6]

    AirPassengers.csv: A classic dataset representing monthly international airline passenger numbers from 1949 to 1960, often used for time series analysis benchmarking due to its clear trend and seasonality [14]

    1949

  7. [7]

    PRSA_data_2010_pm25.csv: Contains hourly PM2.5 air quality data from Beijing, which commonly exhibits missing values due to sensor issues or environmental factors, making it relevant for real-world imputation scenarios [17]

  8. [8]

    opsd_germany_daily.csv: Features daily electricity consumption, wind power production, and solar power production data for Germany, a dataset that can have reporting gaps or periods of no generation for certain sources [42]

Show all 95 references
  1. [9]

    KZImputer

    Additional Dataset Example, e.g., financial data like all_stocks_5yr.csv [40] or specific sensor data like T1.csv [23, 57]. For each dataset used in a specific evaluation, artificial missingness would be introduced. Missing data would typically be generated under a Missing Com...

  2. [10]

    Mean Imputation: replaces missing values with the global mean of the series

  3. [11]

    Median Imputation: replaces missing values with the global median of the series

  4. [12]

    (Note: KZImputer uses this for mid-gaps, so this comparison will be particularly relevant for start/end gaps and overall performance)

    Linear Interpolation: fills missing values using linear interpolation between the two nearest observed points. (Note: KZImputer uses this for mid-gaps, so this comparison will be particularly relevant for start/end gaps and overall performance)

  5. [13]

    Suitable for time series or sequential data

    Forward Fill (ffill): Propagates the last valid observation forward to fill missing values. Suitable for time series or sequential data. Last Observation Carried Forward (LOCF): fills missing values with the last observed value

  6. [14]

    Often used when future values are assumed to influence missing entries

    Backward Fill (bfill): Fills missing values with the next valid observation. Often used when future values are assumed to influence missing entries

  7. [15]

    Captures non-linear trends better than linear interpolation

    Spline Interpolate (spline): Uses smooth polynomial functions to estimate missing values. Captures non-linear trends better than linear interpolation

  8. [16]

    Effective for capturing local patterns in the data

    kNN (k = 5): Imputes missing values based on the average of the nearest 5 neighbors in feature space. Effective for capturing local patterns in the data

  9. [17]

    Useful for multivariate data imputation

    Iterative Imputer: Models each feature with missing values as a function of other features and iteratively refines the estimates. Useful for multivariate data imputation. Implementations of these methods will be sourced from standard Python libraries such as `pandas`, `scikit-...

  10. [18]

    Less sensitive to outliers compared to RMSE

    MAE (Mean Absolute Error): Measures average absolute differences between predicted and actual values, providing intuitive interpretation in original units. Less sensitive to outliers compared to RMSE

  11. [19]

    More sensitive to outliers than MAE

    RMSE (Root Mean Square Error): Square root of average squared prediction errors, heavily penalizing large deviations and expressed in original data units. More sensitive to outliers than MAE

  12. [20]

    Undefined when actual values are zero

    MAPE (Mean Absolute Percentage Error): percentage-based error metric showing average absolute deviation as percentage of actual values, useful for relative performance assessment. Undefined when actual values are zero

  13. [21]

    Typically expressed as percentage for interpretability

    NRMSE (Normalized Root Mean Square Error): RMSE normalized by data range or mean, enabling comparison across different scales and datasets. Typically expressed as percentage for interpretability

  14. [22]

    Higher values indicate better model fit to data

    R² (Coefficient of determination): proportion of variance in dependent variable explained by the model, ranging from 0 to 1. Higher values indicate better model fit to data

  15. [23]

    Square root of JS divergence forms a proper distance metric

    JS_Divergence (Jensen -Shannon divergence): symmetric measure of similarity between two probability distributions, bounded between 0 and 1. Square root of JS divergence forms a proper distance metric

  16. [24]

    Particularly useful for comparing distributions with different supports

    Wasserstein distance: measures minimum cost to transform one probability distribution into another, also known as Earth Mover's Distance. Particularly useful for comparing distributions with different supports

  17. [25]

    Values closer to zero indicate better preservation of correlational structure

    Correlation_Diff: difference in correlation coefficients between two datasets or model predictions, measuring how well relationships between variables are preserved. Values closer to zero indicate better preservation of correlational structure. These metrics will be calculated...

  18. [26]

    Gap’s number

    Results This section presents the results of the comparative evaluation of KZImputer against benchmark methods. Figures illustrate representative cases of KZImputer’s behavior , additional results are provided in the Supplementary Materials and the authors’ repository [49]. It...

  19. [27]

    Key observations are summarized in Tables 2 and 3

    Discussion The results above form the basis for the following discussion of KZImputer’s strengths, limitations, and implications. Key observations are summarized in Tables 2 and 3. Table 2. Discussion summary of KZImputer performance characteristics Aspect Key Observation Impl...

  20. [28]

    Suitable for domains with degraded sensors or telemetry data (e.g., remote sensing, anomaly-prone IoT)

    Robustness to high sparsity Maintains signal fidelity even under extreme missingness (up to 70 – 80%). Suitable for domains with degraded sensors or telemetry data (e.g., remote sensing, anomaly-prone IoT)

  21. [29]

    Ensures dynamic consistency essential for downstream tasks (e.g., forecasting, root cause analysis)

    Preservation of structural dynamics Accurately reconstructs both global trends and local variability due to hybrid smoothing–decomposition architecture. Ensures dynamic consistency essential for downstream tasks (e.g., forecasting, root cause analysis)

  22. [30]

    Optimal for environmental, behavioral, or energy datasets where structural evolution is prominent

    Compatibility with trend-driven data Excels with non -stationary inputs featuring slow drifts or nonlinear regimes, as verified by residual diagnostics and ACF plots. Optimal for environmental, behavioral, or energy datasets where structural evolution is prominent

  23. [31]

    Enables valid application of classical models (e.g., ARIMA, Kalman Filters) that assume stationarity

    Statistical consistency of residuals Post-imputation series exhibits lower autocorrelation and more stable variance. Enables valid application of classical models (e.g., ARIMA, Kalman Filters) that assume stationarity

  24. [32]

    Use simpler methods (e.g., spline, EMA) in low-sparsity unless structural fidelity is mission-critical

    Limitations in low- missingness regimes Performance gains diminish below 40% missingness; computational cost may outweigh benefit. Use simpler methods (e.g., spline, EMA) in low-sparsity unless structural fidelity is mission-critical

  25. [33]

    Enhances quality and stability of learning pipelines via structured, nonparametric imputation groundwork

    Potential for hybrid integration Easily integrable as a modular preprocessing component before ML/DL refinement stages. Enhances quality and stability of learning pipelines via structured, nonparametric imputation groundwork. Table 3. Strengths, limitations, implications for p...

  26. [34]

    The methodology, imple mented in Python, combines localized averaging with interpolation to balance accuracy, computational efficiency, and ease of use

    Conclusion This paper presents KZImputer, an adaptive imputation method for univariate time series designed for short to medium-sized gaps (1-5 points and beyond) with tailored strategies for segments at the start, middle, or end of the series. The methodology, imple mented in...

  27. [35]

    Adebayo, Patrick and I, Ahmed. and K. T., Oyeleke, Comparative Analysis of Prophet and ARIMA Models for Forecasting Economic Indicators: A Case Study of Personal Consumption Expenditures (April 05, 2025). Available at SSRN: https://ssrn.com/abstract=5206738

    2025

  28. [36]

    (2021), Kalman filter algorithm versus other methods of estimating missing values: time series evidence

    Adejumo O.A., Onifade O.C., Albert S. (2021), Kalman filter algorithm versus other methods of estimating missing values: time series evidence. African Journal of Mathematics and Statistics Studies 4(2), 1 -9. DOI: 10.52589/AJMSSVFVNMQLX

    2021 doi

  29. [37]

    and Miskon, S

    Adnan, F.A., Jamaludin, K.R., Wan Muhamad, W.Z.A. and Miskon, S. (2022) A Review of the Current Publication Trends on Missing Data Imputation over Three Decades: Direction and Future Research. Neural Computing and Applications, 34, 18325-18340. https://doi.org/10.1007/s00521-0...

    2022 doi

  30. [38]

    Conceptual framework as a guide to choose appropriate imputation method for missing values in a clinical structured dataset

    Afkanpour M, Tehrany Dehkordy D, Momeni M, Tabesh H. Conceptual framework as a guide to choose appropriate imputation method for missing values in a clinical structured dataset. BMC Med Res Methodol. 2025 Feb 20;25(1):43. doi: 10.1186/s12874-025-02496-3

    2025 doi

  31. [39]

    Al-Helali, B., Chen, Q., Xue, B. et al. A new imputation method based on genetic programming and weighted KNN for symbolic regression with incomplete data. Soft Comput 25, 5993–6012 (2021). https://doi.org/10.1007/s00500-021- 05590-y

    2021 doi

  32. [40]

    and Pinto, F.C

    Almeida, A., Brás, S., Sargento, S. and Pinto, F.C. (2024) Focalize K-NN: An Imputation Algorithm for Time Series Datasets. Pattern Analysis and Applications, 27, Article No. 39. https://doi.org/10.1007/s10044-024-01262-3

    2024 doi

  33. [41]

    G., & Bland, J

    Altmann, D. G., & Bland, J. M. (2007). Missing data. British Medical Journal , 334(7590), 424. DOI: 10.1136/bmj.38977.682025.2C

    2007

  34. [42]

    and Gad, I

    Alwateer, M., Atlam, E., El -Raouf, M., Ghoneim, O. and Gad, I. (2024) Missing Data Imputation: A Comprehensive Review. Journal of Computer and Communications, 12, 53-75. doi: 10.4236/jcc.2024.1211004

    2024

  35. [43]

    A Review of Hot Deck Imputation for Survey Non -response

    Andridge RR, Little RJ. A Review of Hot Deck Imputation for Survey Non -response. Int Stat Rev. 2010 Apr;78(1):40-64. doi: 10.1111/j.1751-5823.2010.00103.x. PMID: 21743766; PMCID: PMC3130338

    2010

  36. [46]

    J., Stuart, E

    Azur, M. J., Stuart, E. A., Frangakis, C., & Leaf, P. J. (2011). Multiple imputation by chained equations: What is it and how does it work? International Journal of Methods in Psychiatric Research, 20(1), 40–49. DOI: 10.1002/mpr.329

    2011 doi

  37. [47]

    W., & Hatch, L

    Beck, M. W., & Hatch, L. K. (2023). A review of machine -learning approaches to missing -data imputation in environmental sciences. Environmental Modelling & Software, 160, 105631. https://doi.org/10.1016/j.envsoft.2023.105631

    2023

  38. [48]

    Box, G. E. P., Jenkins, G. M. and Reinsel, G. C. (1976) Time Series Analysis, Forecasting and Control. Third Edition. Holden-Day. Series G

    1976

  39. [49]

    Cao, W., Wang, D., Li, J., Zhou, H., Li, L., & Li, Y . (2018). BRITS: Bidirectional recurrent imputation for time series. In Advances in Neural Information Processing Systems 31 (pp. 6776 –6786). https://doi.org/10.48550/arXiv.1805.10572

    2018 doi arXiv

  40. [50]

    Casella M, Milano N, Dolce P and Marocco D (2024) Transformers deep learning models for missing data imputation: an application of the ReMasker model on a psychometric scale. Front. Psychol . 15:1449272. doi: 10.3389/fpsyg.2024.1449272

    2024

  41. [51]

    Chang Wei Tan, Christoph Bergmeir, Francois Petitjean, & Geoffrey I Webb. (2020). Beijing PM2.5 Dataset [Data set]. Zenodo. https://doi.org/10.5281/zenodo.3902671

    2020 doi

  42. [52]

    Chen, L., Wang, Y ., & Zhang, M. (2017). Multiple imputation of missing wind power data using Bayesian networks. Journal of Renewable and Sustainable Energy, 9(2), 023304. https://doi.org/10.1063/1.4977309

    2017 doi

  43. [53]

    Chen, Mei & Zhu, Hongyu & Chen, Yongxu & Wang, Youshuai. (2022). A Novel Missing Data Imputation Approach for Time Series Air Quality Data Based on Logistic Regression. Atmosphere. 13. 1044. 10.3390/atmos13071044

    2022 doi

  44. [54]

    (2023) Comparison of Imputation Methods for Univariate Time Series

    Chhabra, G. (2023) Comparison of Imputation Methods for Univariate Time Series. International Journal on Recent and Innovation Trends in Computing and Communication , 11, 286 -292. https://doi.org/10.17762/ijritcc.v11i2s.6148

    2023 doi

  45. [55]

    and Tabona, O

    Emmanuel, T., Maupong, T., Mpoeleng, D., Semong, T., Mphago, B. and Tabona, O. (2021) A Survey on Missing Data in Machine Learning. Journal of Big Data , 8, Article No. 140. https://doi.org/10.1186/s40537-021-00516-9

    2021 doi

  46. [56]

    (2022) Applied Missing Data Analysis

    Enders, C.K. (2022) Applied Missing Data Analysis. Guilford Publications

    2022

  47. [57]

    EMHIRES dataset

    Gonzalez, A.I; Zucker, A.; Careri, F.; Monforti, F.; Huld, T.; Badger, J. EMHIRES dataset. Part I: Wind power generation European Meteorological derived HIgh resolution RES generation time series for present and future scenarios. (2016) EUR 28171 EN; https://doi.org/10.2790/83...

    2016 doi

  48. [59]

    Accuracy of random -forest-based imputation of missing data in the presence of non - normality, non-linearity, and interaction

    Hong, S., Lynn, H.S. Accuracy of random -forest-based imputation of missing data in the presence of non - normality, non-linearity, and interaction. BMC Med Res Methodol 20, 199 (2020). https://doi.org/10.1186/s12874-020-01080- 1

    2020 doi

  49. [60]

    Jahangiri, M., Kazemnejad, A., Goldfeld, K.S. et al. A wide range of missing imputation approaches in longitudinal data: a simulation study and real data analysis. BMC Med Res Methodol 23, 161 (2023). https://doi.org/10.1186/s12874-023-01968-8

    2023 doi

  50. [61]

    Jha, S., Wang, J., Marina, N. (2022). A novel model-driven deterministic approach to wind power imputation. Sustainable Computing: Informatics and Systems. 36. 100818. https://doi.org/10.1016/j.suscom.2022.100818

    2022

  51. [62]

    Kavun S., Zamula, A., Miziurin, V . Intelligent Evaluation Method for Complex Systems in The Big Data Environment, 2019 IEEE 2nd Ukraine Conference on Electrical and Computer Engineering (UKRCON), July 2 -6, Lviv, Ukraine, 2019, pp. 951-957. https://doi.org/10.1109/ UKRCON.201...

    2019

  52. [63]

    Conceptual fundamentals of a theory of mathematical interpretation

    Kavun, S. Conceptual fundamentals of a theory of mathematical interpretation. Int. J. Computing Science and Mathematics 2015, V olume 6, No. 2, pp. 107–121. https://doi.org/10.1504/IJCSM.2015.069459

    2015

  53. [64]

    (2021) Addressing Missing Data in Educational Evaluation: Predictive Mean Matching Imputation for Test Score Data

    Khanani, N. (2021) Addressing Missing Data in Educational Evaluation: Predictive Mean Matching Imputation for Test Score Data. Proceedings of the 2021 AERA Annual Meeting , 4 September 2021, 15. https://doi.org/10.3102/1687298

    2021 doi

  54. [65]

    Exponential moving average versus moving exponential average

    Klinker, F. Exponential moving average versus moving exponential average. Math Semesterber 58, 97–107 (2011). https://doi.org/10.1007/s00591-010-0080-8

    2011 doi

  55. [66]

    Kotenko, S., Nitsenko, V ., Hanzhurenko, I., & Havrysh, V . (2020). The mathematical modeling stages of combining the carriage of goods for indefinite, fuzzy and stochastic parameters. International Journal of Integrated Engineering, 12(7), 173-180. https://doi.org/10.30880/ij...

    2020 doi

  56. [67]

    and Samad, M.D

    Kowsar, I., Rabbani, S.B. and Samad, M.D. (2024) Attention -Based Imputation of Missing Values in Electronic Health Records Tabular Data. 2024 IEEE 12th International Conference on Healthcare Informatics (ICHI), Orlando, 3-6 June 2024, 177-182. https://doi.org/10.1109/ichi6124...

    2024

  57. [68]

    H., Park, Y

    Lee, S. H., Park, Y . H., & Kim, W. (2019). Temporal pattern-based imputation for wind energy datasets. IEEE Transactions on Sustainable Energy, 10(2), 865-875. https://doi.org/10.1109/TSTE.2018.2852093

    2019

  58. [69]

    Little, R. J. A., & Rubin, D. B. (2019). Statistical analysis with missing data (3rd ed.). Wiley. https://doi.org/10.1002/9781119482260

    2019 doi

  59. [70]

    A statistical learning framework for the intelligent imputation of offshore wind farm missing SCADA data

    Liu, X.; Zheng, Z.; Zhang, Z.; Cao, Z. A statistical learning framework for the intelligent imputation of offshore wind farm missing SCADA data. Proceedings of the 8th Renewable Power Generation Conference (RPG 2019), Shanghai, China, 24-25 Oct. 2019. https://doi.org/10.1049/c...

    2019

  60. [71]

    Liu, Z., Jiang, H., & Du, P. (2016). Missing data imputation for wind speed using machine learning methods. Energy Procedia, 100, 428-433. https://doi.org/10.1016/j.egypro.2016.10.157

    2016 doi

  61. [72]

    C.P.; Jiang, F.; Chen, W.; Wang, M.; Zhai

    Ma, J.; Cheng, J. C.P.; Jiang, F.; Chen, W.; Wang, M.; Zhai. C. A bi -directional missing data imputation scheme based on LSTM and transfer learning for building energy data. Energy and Buildings 2020, Volume 216, Article 109999. https://doi.org/10.1016/j.enbuild.2020.109941

    2020

  62. [73]

    and Akbar, M.I

    Murti, D.M.P., Pujianto, U., Wibawa, A.P. and Akbar, M.I. (2019) K -Nearest Neighbor (K -NN) Based Missing Data Imputation. 2019 5 th International Conference on Science in Information Technology (ICSITech), Yogyakarta, 23-24 October 2019, 83-88. https://doi.org/10.1109/icsite...

    2019

  63. [74]

    Nitiraj Kulkarni, & Jagadish Tawade. (2024). Dataset: iShares 1 -5 Year Investment Grade Corporate Bond ETF (IGSB) Stock Performance [Data set]. Zenodo. https://doi.org/10.5281/zenodo.12558266

    2024 doi

  64. [75]

    Nitsenko, V ., Kotenko, S., Hanzhurenko, I., Mardani, A., Stashkevych, I., Karakai, M. (2020). Mathematical Modeling of Multimodal Transportation Risks. In: Ghazali, R., Nawi, N., Deris, M., Abawajy, J. (eds) Recent Advances on Soft Computing and Data Minin g. SCDM 2020. Advan...

    2020 doi

  65. [76]

    Open Power System Data. 2020. Data Package Time series. Version 2020 -10-06. https://doi.org/10.25832/time_series/2020-10-06

    2020 doi

  66. [77]

    , Mikheev, I

    Panchenko, V ., Zamula, A., Kavun, S. , Mikheev, I. Intelligent management of the enterprise personnel security system. 2018 IEEE 9th International Conference on Dependable Systems, Services and Technologies (DESSERT), Kiev, 2018, pp. 469-474, https://doi.org/10.1109/DESSERT.2...

    2018

  67. [78]

    and Askounis, D

    Psychogyios, K., Ilias, L., Ntanos, C. and Askounis, D. (2023) Missing Value Imputation Methods for Electronic Health Records. IEEE Access, 11, 21562-21574. https://doi.org/10.1109/access.2023.3251919

    2023

  68. [79]

    A novel wind turbine data imputation method with multiple optimizations based on GANs

    Qu, F.; Liu, J.; Ma, Y .; Zang, D.; Fu, M. A novel wind turbine data imputation method with multiple optimizations based on GANs. Mechanical Systems and Signal Processing 2020, Volume 139 , Article 106610. https://doi.org/10.1016/j.ymssp.2019.106610

    2020

  69. [80]

    Rubin, D. B. (1976). Inference and missing data. Biometrika, 63(3), 581 –592. https://doi.org/10.1093/biomet/63.3.581

    1976 doi

  70. [81]

    Russi, L., Guidorzi, P., Pulvirenti, B., Aguiari, D., Pau, G., & Semprini, G. (2022). Air Quality and Comfort Characterisation within an Electric Vehicle Cabin in Heating and Cooling Operations. Sensors, 22(2), 543. https://doi.org/10.3390/s22020543

    2022 doi

  71. [82]

    Simulation of exchange processes in multi-component environments with account of data uncertainty

    S. Gadetska, V . Dubnitskiy, Y . Kushneruk, Y . Ponochovnyi, and A. Khodyrev, “Simulation of exchange processes in multi-component environments with account of data uncertainty”, Advanced Information Systems , vol. 8, no. 1, pp. 12–23, 2024, doi: https://doi.org/10.20998/2522-...

    2024

  72. [83]

    Sergii Kavun. (2025). s -kav/kz_data_imputation: KZImputer: Implementation of the KZImputer. [KZImputer class]. Zenodo. https://doi.org/10.5281/zenodo.15663430

    2025 doi

  73. [84]

    and Trubitsyna, I

    Shahbazian, R. and Trubitsyna, I. (2022) DEGAIN: Generative -Adversarial-Network-Based Missing Data Imputation. Information, 13, Article 575. https://doi.org/10.3390/info13120575

    2022 doi

  74. [85]

    Blomberg, Orlin S

    Simone P. Blomberg, Orlin S. Todorov. The fallacy of single imputation for trait databases: Use multiple imputation instead, Methods in Ecology and Evolution, DOI: 10.1111/2041-210X.14494, 16, 4, (658-667), (2025)

    2025 doi

  75. [86]

    Tawn, R., Browell, J., Dinwoodie, I. (2020). Missing data in wind farm time series: Properties and effect on forecasts, vol. 189, 106640. https://doi.org/ 10.1016/j.epsr.2020.106640

    2020

  76. [87]

    (2024) Applications: Imputation of Missing Hydrometeorological Data

    Teegavarapu, R.S.V . (2024) Applications: Imputation of Missing Hydrometeorological Data. Water Science and Technology Library, 108, 491-517. https://doi.org/10.1007/978-3-031-60946-6_8

    2024 doi

  77. [88]

    and Singh, D.P

    Thakur, S., Choudhary, J. and Singh, D.P. (2021) A Survey on Missing Values Handling Methods for Time Series Data. In: Sheth, A., Sinhal, A., Shrivastava, A. and Pandey, A.K., Eds., Intelligent Systems, Springer Singapore, 435 -

    2021

  78. [90]

    van Buuren, S. (2018). Flexible Imputation of Missing Data, Second Edition (2nd ed.). Chapman and Hall/CRC. https://doi.org/10.1201/9780429492259

    2018 doi

  79. [91]

    Wei Cao, Dong Wang, Jian Li, Hao Zhou, Yitan Li, and Lei Li. 2018. BRITS: bidirectional recurrent imputation for time series. In Proceedings of the 32nd International Conference on Neural Information Processing Systems (NIPS'18). Curran Associates Inc., Red Hook, NY , USA, 6776–6786

    2018

  80. [92]

    2018 Scada Data of a Wind Turbine in Turkey

    Wind Turbine Scada Dataset. 2018 Scada Data of a Wind Turbine in Turkey. Dataset owner: Berk Erisen. Available: https://www.kaggle.com/berkerisen/wind-turbine-scada-dataset

    2018

  81. [93]

    E., Halvorson, M., King, K

    Woods, A.D., Gerasimova, D., Van Dusen, B., Nissen, J., Bainter, S., Uzdavines, A., Davis-Kean, P. E., Halvorson, M., King, K. M., Logan, J. A. R., Xu, M., Vasilev, M. R., Clay, J. M., Moreau, D., Joyal-Desmarais, K., Cruz, R. A., Brown, D. M. Y ., Schmidt, K., & Elsherif, M. ...

    2024 doi

  82. [94]

    A Graph-Based Approach for Missing Sensor Data Imputation,

    X. Jiang, Z. Tian and K. Li, "A Graph-Based Approach for Missing Sensor Data Imputation," in IEEE Sensors Journal, vol. 21, no. 20, pp. 23133-23144, 15 Oct.15, 2021, doi: 10.1109/JSEN.2021.3106656

    2021

  83. [95]

    Yusuke Tashiro, Jiaming Song, Yang Song, and Stefano Ermon. 2021. CSDI: conditional score -based diffusion models for probabilistic time series imputation. In Proceedings of the 35th International Conference on Neural Information Processing Systems (NIPS '21). Curran Associate...

    2021

  84. [96]

    Complex systems modeling with intelligent control elements

    Zamula, A., Kavun, S. Complex systems modeling with intelligent control elements. International Journal of Modeling, Simulation, and Scientific Computing 2017, V olume 8, No. 1. https://doi.org/10.1142/S179396231750009X

    2017 doi

  85. [97]

    Zamula, A., Kavun, S., Serdukov, K. Binary Recommender System with Artificial Intelligence Aids, 2019 IEEE International Scientific -Practical Conference Problems of Infocommunications, Science and Technology (PIC S&T), Kharkiv, Ukraine, 2019, pp. 251-255. https://doi.org/10.1...

    2019

  86. [98]

    Zhang, Q., Wang, C., & Liu, Y . (2020). Deep learning-based imputation for missing data in wind energy time series. Renewable Energy, 147(Part 1), 872-882. https://doi.org/10.1016/j.renene.2019.09.044

    2020 doi

  87. [443]

    https://doi.org/10.1007/978-981-16-2248-9_42

    doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.