Pith. sign in

REVIEW 3 major objections 6 minor 29 references

Anomaly Detection in High Dimensional Data

T0 review · 3 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The stray algorithm turns the largest k-nearest-neighbour gap into one anomaly score that catches singletons, micro clusters, and inliers, without the clustering step that slows and misleads HDoutliers.

desk verdict Useful incremental algorithm with a simple max-gap kNN score, but the EVT threshold is applied to dependent scores without validation and the reported FPR doesn't match the nominal α; worth engaging if the threshold step is fixed. read the letter →

arxiv 1908.04000 v1 pith:6KWP6NHW submitted 2019-08-12 stat.ML cs.LGstat.AP

Priyanga Dilini Talagala , Rob J. Hyndman , Kate Smith-Miles This is my paper · ORCID
classification stat.MLcs.LGstat.AP MSC 62H3062G32
keywords anomalydetectionhigh-dimensionaldatak-nearestneighboursmaximumgapextremevaluetheorymicroclustersinliersstrayalgorithm
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a single distance-based score—the k-nearest-neighbour distance at the largest gap in the k-nearest-neighbour distance curve—can detect the anomalies that the HDoutliers algorithm misses, without the clustering pass that HDoutliers uses. The score makes anomalous singletons, micro clusters whose size is below k, and inliers between dense typical clusters visible, and it feeds a data-driven extreme-value threshold that separates typical points from anomalies. The paper shows on a series of counterexamples that this stray algorithm matches or beats both versions of HDoutliers in accuracy and is substantially faster, and it demonstrates the same approach on pedestrian-count data after converting days and sensors into feature vectors. If the claim holds, high-dimensional anomaly detection in fraud, security, environmental monitoring, and urban planning can be done with a simpler, faster, and more sensitive tool.

What carries the argument

The carrying object is the k-nearest-neighbour distance with the maximum gap, defined for each point as follows: compute its distances to the k nearest neighbours, take the successive differences between ordered distances, and record the largest difference as the point's anomalous score. For a singleton anomaly the maximum gap is typically the first nearest-neighbour distance; for a micro cluster of size less than k it is the jump from the cluster to the main body; for an inlier it is the large gap that separates it from every dense class. This score replaces the Leader-algorithm clustering step in HDoutliers, so the algorithm never down-samples the data and never lets close anomalous clusters act as each other's nearest neighbours. The threshold is set by assuming the scores' upper tail follows the exponential spacing law that governs extremes from the Gumbel maximum domain of attraction (a broad class of light-to-moderate-tailed distributions): the algorithm starts from the 50% of points with the smallest scores, fits an exponential to their upper tail, and walks upward through the remaining scores, updating the cut-off until a score exceeds it. The bottom-up search yields a binary typical/anomalous classification, and the scores themselves give each point a continuous anomaly rank.

What would settle it

Run the stray algorithm with α = 0.05 on anomaly-free data drawn from a heavy-tailed multivariate distribution (for example, a multivariate t with three degrees of freedom) across several dimensions and sample sizes; if the empirical false-positive rate clearly exceeds 0.05, or if the upper-tail spacings of the max-gap scores fail an exponential fit, the threshold assumption is not doing the claimed work.

Watch

Extended reading notes

Core claim

The central claim is that an anomaly is best characterised as a point whose nearest-neighbour structure contains a large distance gap, and that the largest such gap among the k nearest neighbours is a sufficient anomalous score in any dimension. On this definition, a group of up to k anomalous points no longer masks itself: each member sees the others among its nearest neighbours, but the jump from the group to the main body of data produces the maximum gap, so the whole group is flagged. The same score also catches inliers sitting between two dense classes, because their nearest neighbours are far away compared with the gaps inside the classes. The stray algorithm computes this score for every point, derives a cut-off from an extreme-value spacing model of the upper tail of the scores, and classifies everything above the cut-off as anomalous, assigning each point a continuous outlier rank. The paper demonstrates, on bivariate counterexamples and on the Melbourne pedestrian data, that this design removes the false negatives and false positives caused by the Leader-algorithm clustering step in HDoutliers and runs faster at scale.

Load-bearing premise

The load-bearing premise is that the largest k-nearest-neighbour gap scores follow the same exponential spacing law as extremes of independent random values, so the extreme-value threshold carries its claimed false-positive probability; on dependent scores, or scores whose upper tail is not Gumbel-domain, the cut-off would be miscalibrated.

Editorial extensions

If this is right

  • Micro clusters of up to k points are detected directly from the gap score, so two anomalous clusters that sit close together no longer hide each other the way they do under HDoutliers.
  • Inliers—points that fall between dense typical classes—are flagged as anomalies, giving the algorithm a capability that the HDoutliers variants fail on when the inliers are near one another.
  • At α = 0.05 on anomaly-free Gaussian data, the stray algorithm reports lower false positive rates than both HDoutliers versions across dimensions 1, 10, and 100 and sample sizes from 100 to 10,000.
  • Because no clustering step is needed, stray runs far more quickly than HDoutliers with clustering, especially in dimensions above 10 and for large samples, and its brute-force variant is competitive with k-d-tree search.
  • The same algorithm carries over to time series and collections of scatterplots after feature engineering, which the paper demonstrates by detecting New Year's Eve and a concert day in Melbourne pedestrian data and by isolating the Southbank sensor.

Reading between the lines

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

  • Beyond the paper's examples, the same maximum-gap score could be applied to any distance matrix—kernel distances, graph distances, or embeddings—so the method may transfer to non-Euclidean anomaly detection after the same min-max normalisation.
  • A testable next step is to choose k from the data instead of by hand: since the paper interprets k as the maximum permissible size of a micro cluster, an adaptive k tied to local density or to the distribution of nearest-neighbour distances would remove the main user-set parameter.
  • If the extreme-value threshold is approximately correct for a wider class of distributions, the same spacing-based cutoff could give probabilistic meaning to the outputs of other distance- and density-based outlier scores, which usually rely on user-chosen thresholds.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes the 'stray' algorithm for unsupervised anomaly detection in high-dimensional data. Each observation is scored by the maximum gap among its k-nearest-neighbour distances, and a binary classification is obtained by comparing this score with a threshold derived from extreme value theory (EVT), following the approach of Weissman's spacing theorem as used in HDoutliers. The authors argue that this score addresses three limitations of HDoutliers: masking by nearby anomalous clusters, information loss from the Leader clustering step, and threshold instability. They illustrate the method on synthetic counterexamples, compare running times and false positive rates against HDoutliers, and demonstrate an application to pedestrian-count data using feature engineering. An R package, stray, is provided.

Significance. If the claims were fully supported, the paper would offer a useful, fast, clustering-free anomaly detector that handles singletons, micro clusters, and inliers, with an open-source implementation. The strengths of the manuscript include a clearly specified algorithm, an R package that makes the method reproducible, and an explicit comparison with a specific baseline algorithm (HDoutliers) on targeted examples that expose known failure modes. The paper also honestly acknowledges in Section 7 that the counterexamples are not diverse enough to claim general superiority. However, the central claim of a 'valid probabilistic interpretation' for the threshold rests on an EVT assumption that is not justified for the proposed score, and the empirical calibration in Table 1 appears to contradict the nominal error rate. The value of the paper therefore depends on whether the threshold step can be theoretically or empirically supported.

major comments (3)
  1. [Section 4.4, Theorem 4.1] Weissman's spacing theorem is stated for spacings of order statistics from an iid sample whose distribution lies in the maximum domain of attraction of the Gumbel distribution. The paper applies this theorem directly to the kNN max-gap scores, but these scores are neither independent nor identically distributed: each score is a maximum gap among k nearest-neighbour distances, and points share neighbours; in multimodal data the score distribution is a mixture; and under min-max normalization the scores have bounded support in [0, 1], whose domain-of-attraction class is not established. Consequently, the 'valid probabilistic interpretation' claimed as an advantage in Section 1 (advantage 3) is unsupported. The authors should either provide a rigorous justification, or present simulation evidence that the resulting threshold is calibrated across distributions, dimensions, and sample sizes, or explicitly rescope the threshold as a heuristic.
  2. [Table 1] Table 1 reports false positive rates for stray of approximately 0.006 for d=1 with alpha=0.05, falling to 0.000 or 0.001 for higher dimensions. If the threshold is intended to be the upper 1-alpha quantile of the fitted exponential distribution, the achieved FPR should be close to 0.05, not nearly an order of magnitude smaller. The text frames the low FPR as a positive feature, but this is in tension with the claimed probabilistic interpretation of the threshold. The authors should report the achieved FPR as a function of alpha for several configurations and reconcile the discrepancy, or revise the claim that the threshold has a nominal alpha interpretation.
  3. [Section 4.4, bottom-up procedure] The threshold calculation fits an exponential distribution to the 'upper tail' of the lower 50% of the scores, i.e., around the median of the score distribution, rather than to the extreme upper tail where Weissman's spacing theorem operates. The subsequent bottom-up updating, which adds one point at a time and recomputes the cutoff, moves even further from a fixed EVT quantile. The procedure may work as a heuristic, but it should not be described as an EVT-based quantile without additional theoretical or empirical support.
minor comments (6)
  1. [Section 5] The text says the values in Table 1 are based on 1000 iterations, while the table caption says 100 iterations; this inconsistency should be corrected.
  2. [Section 4.4, Theorem 4.1] The definition D_{i,n} = X_{i:n} - X_{i+1:n} with ascending order statistics gives non-positive spacings; the theorem should be stated for upper-order spacings, e.g., D_{i,n} = X_{n-i+1:n} - X_{n-i:n}, to match Figure 4 and the standard Weissman formulation.
  3. [Section 4.3] In the sentence 'The second dataset, in Figure 3 b), has three anomalies', the reference should likely be Figure 3(c) or Figure 3(d); please correct.
  4. [Section 4.3] The sentence 'the k-nearest neighbour distance with the maximum gap occurs when k = 1' appears to mean the first gap (i = 1) is the maximum; please clarify the notation.
  5. [Section 4.2] The claim that min-max normalization 'does not change the distribution or squeeze points together masking anomalies' is too strong: linear rescaling can alter distances when variables have very different scales, and the presence of anomalies can affect the min-max range. Please soften or provide a supporting reference.
  6. [Section 4] A pseudo-code block or algorithm box summarizing the full stray procedure would improve clarity, since the score computation (Section 4.3) and the threshold computation (Section 4.4) are described in separate subsections.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the kNN max-gap score and the EVT threshold are defined independently of the anomaly labels, and the comparison with HDoutliers is empirical rather than derived from the algorithm's own construction.

full rationale

The paper's central claim is that the stray algorithm, using the k-nearest-neighbour distance at the maximum gap as an anomaly score, overcomes specific failure modes of HDoutliers. I examined the derivation chain for any step where an output is equivalent to an input by construction, or where a fitted parameter is renamed as a prediction, or where a load-bearing premise rests on a self-citation. No such step exists. The anomaly score in Section 4.3 is defined purely from pairwise distances: for each point the algorithm computes k nearest-neighbour distances, takes successive differences, and selects the distance at which the maximum gap occurs. This definition is independent of class labels, of the HDoutliers output, and of the threshold calculation; it is a heuristic operationalization of the paper's stated definition of an anomaly as an observation separated by a large distance gap. The threshold in Section 4.4 is not fitted to known anomalies. It follows Weissman's spacing theorem as cited to Weissman (1978), Burridge and Taylor (2006), Schwarz (2008), and Wilkinson (2017), and it uses only the lower half of the scores to fit an exponential tail. The subsequent classification is an exceedance of that data-driven threshold, which is standard unsupervised thresholding rather than a prediction forced by a fitted input. The paper does contain self-citations, for example Talagala et al. (2019a, 2019b) and Hyndman, Wang and Laptev (2015), but these are used for background, feature construction, or R package implementation; they are not load-bearing for the central algorithm or for the claim that stray outperforms HDoutliers on the illustrated counterexamples. The main weakness of the paper is that the applicability of the EVT assumption to the kNN max-gap score is asserted rather than proved, and Table 1 reports false-positive rates far below the nominal alpha = 0.05. That is a correctness or calibration concern, not a circularity concern. Consequently, the appropriate circularity finding is no significant circularity, with a score of 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a distance-gap definition of anomalies, a user-set neighborhood size k, and an unverified extension of Weissman's spacing theorem to dependent kNN max-gap scores. No new physical or mathematical entities are introduced.

free parameters (3)
  • k (number of nearest neighbors) = 10 in micro-cluster examples; user-set
    Section 4.3 introduces k as a user-defined parameter with no automatic selection or sensitivity analysis, despite the algorithm's performance depending on it.
  • alpha (critical value) = 0.05 in experiments
    Section 5 uses alpha=0.05 for false positive rate evaluation; the threshold depends on this choice.
  • initial typical fraction (0.5) = 0.5
    Section 4.4 starts the bottom-up search with the 50% smallest anomaly scores, assuming they are all typical; this fixed fraction is a hand-chosen model constant.
assumptions (5)
  • ad hoc to paper An anomaly is an observation that deviates markedly from the majority with a large distance gap, i.e., the gap between typical points and anomalies is larger than gaps among typical points.
    Section 2.2 defines anomalies this way; the algorithm's score is built around this definition, so it can only detect distance-gap anomalies, not density-based or pattern-based anomalies.
  • domain assumption The distribution of the 'k-nearest neighbour distance with maximum gap' scores is in the maximum domain of attraction of the Gumbel distribution.
    Stated in Section 4.4 as an assumption; no evidence or derivation is provided. If false, the EVT threshold is invalid.
  • domain assumption Weissman's spacing theorem applies to the dependent, non-iid kNN max-gap scores as though they were spacings of iid order statistics.
    Section 4.4 applies the theorem to anomalous scores; scores across points are dependent, and each score is a max over gaps, not an order statistic of a simple sample.
  • domain assumption The 50% of points with the smallest anomaly scores contain only typical points, so the fitted exponential tail is uncontaminated.
    Section 4.4 bottom-up search begins with this subset; if anomalies exceed 50% or overlap the smallest scores, the threshold is biased.
  • domain assumption Min-max normalization to the unit hypercube makes Euclidean distances comparable across dimensions.
    Section 4.2 uses this normalization; it is standard but can distort distances for skewed data or outliers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Anomaly Detection in High Dimensional Data." pith.science (2026). https://pith.science/paper/6KWP6NHW

@misc{pith2026190804000,
  author       = {Pith},
  title        = {Pith review of: Anomaly Detection in High Dimensional Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6KWP6NHW}},
  note         = {Machine review of arXiv:1908.04000}
}
read the original abstract

The HDoutliers algorithm is a powerful unsupervised algorithm for detecting anomalies in high-dimensional data, with a strong theoretical foundation. However, it suffers from some limitations that significantly hinder its performance level, under certain circumstances. In this article, we propose an algorithm that addresses these limitations. We define an anomaly as an observation that deviates markedly from the majority with a large distance gap. An approach based on extreme value theory is used for the anomalous threshold calculation. Using various synthetic and real datasets, we demonstrate the wide applicability and usefulness of our algorithm, which we call the stray algorithm. We also demonstrate how this algorithm can assist in detecting anomalies present in other data structures using feature engineering. We show the situations where the stray algorithm outperforms the HDoutliers algorithm both in accuracy and computational time. This framework is implemented in the open source R package stray.

Figures

Figures reproduced from arXiv: 1908.04000 by the authors.

Figure 1
Figure 1. Different types of anomalies in high-dimensional data. Anomalies are represented by red triangles and black dots correspond to the typical behaviour. high-dimensional space and then selecting a representative member from each cluster before calculating nearest neighbour distances for the selected instances. In this study, we focus on all three of these anomaly types. 2.2 Definitions for Anomalies in High Dimensional… view at source ↗
Figure 5
Figure 5. Scalability Performance. (a) HDoutliers algorithm without clustering step, (b-I) HDoutliers algorithm with clustering step, (c) stray algorithm with brute force nearest neighboursearch using FNN R package implementation, (d) stray algorithm with kd-trees nearest neighbour search using ‘FNN’ R package implementation, (e) stray algorithm with brute force nearest neighbour search using ‘nabor’ R package implementation,… view at source ↗
Figure 6
Figure 6. Algorithm performance. (a) The top panel shows the results of the HDoutliers algorithm without a clustering step. (b) The middle panel shows the results of the HDoutliers algorithm with a clustering step. The representative member selected from each cluster formed by the Leader algorithm are marked in blue colour. (c) The bottom panel shows the results of the improved algorithmwith bruteforce k-nearest neighboursear… view at source ↗
Figures from the paper (2 more)
Figure 8
Figure 8. Figure 8: Feature -based representation of the collection of scatterplots using Scagnostics. In each plot anomalies determined by the stray algorithm are represented by red colour. to Talagala et al. 2019a; Hyndman, Wang & Laptev 2015) and convert the original collectionof time …
Figure 10
Figure 10. Figure 10: Feature-based representation of the collection of time series on 31 December 2018. In each plot, anomalies determined by the stray algorithm are represented in red colour demonstrate how the stray algorithm can assist in detecting anomalies present in other data struc…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    Anomaly Detection in High Dimensional Data Priyanga Dilini Talagala Department of Econometrics and Business Statistics, Monash University, Australia, and ARC Centre of Excellence for Mathematics and Statistical Frontiers Email: dilini.talagala@monash.edu Corresponding author Rob J. Hyndman Department of Econometrics and Business Statistics, Monash Univers...

    work page 2019

  2. [2]

    Burridge, P & AMR Taylor (2006)

    ACM, pp.93–104. Burridge, P & AMR Taylor (2006). Additive outlier detection via extreme-value theory. Journal of Time Series Analysis 27(5), 685–701. Campos, GO, A Zimek, J Sander, RJ Campello, B Micenková, E Schubert, I Assent & ME Houle (2016). On the evaluation of unsupervised outlier detection: measures, datasets, and an empirical study. Data Mining a...

    work page 2006

  3. [5]

    Anomaly Detection in High Dimensional Data Talagala, Hyndman, Smith-Miles: 12 August 2019 17 The HDoutliers algorithm has two versions

    was designed to test the effect of the dimension, size of the data and the k-nearest neighbour searching method on running times of the different versions of the two algorithms: stray and HDoutliers. Anomaly Detection in High Dimensional Data Talagala, Hyndman, Smith-Miles: 12 August 2019 17 The HDoutliers algorithm has two versions. The first version cal...

    work page 2019

  4. [7]

    Anomalies are represented by red triangles and black dots correspond to the typical behaviour

    addresses this problem to some extent by grouping instances together that are very close in the Anomaly Detection in High Dimensional Data Talagala, Hyndman, Smith-Miles: 12 August 2019 6 Figure 1: Different types of anomalies in high-dimensional data. Anomalies are represented by red triangles and black dots correspond to the typical behaviour. high-dime...

    work page 1969

  5. [9]

    This point corresponds to the sensor positions at Southbank in Melbourne where New Year’s Eve fireworks attract millions of spectators annually

    The stray algorithm declares one point as an anomalous point in this high dimensional data space. This point corresponds to the sensor positions at Southbank in Melbourne where New Year’s Eve fireworks attract millions of spectators annually. These types of findings play a critical role to make decisions about urban planning and man- agement; to identify ...

    work page 2019

  6. [10]

    The focus is then to detect anomalous instances (rows) in the dataset

    numerical variables, where each column corresponds to an attribute and each row corresponds to an observation of these attributes. The focus is then to detect anomalous instances (rows) in the dataset. Anomaly Detection in High Dimensional Data Talagala, Hyndman, Smith-Miles: 12 August 2019 11 4.2 Normalise the Columns Since the stray algorithm is based o...

    work page 2019

  7. [11]

    The second dataset, in Figure 3 b), has three anomalies around (15, 16.5). If we calculate only the nearest neighbour distances for each observation, then the three anomalies are not distinguishable from the typical points since their values are very small (0.7) compared with that of most typical points with nearest neighbour distances at around (0.0015 t...

    work page 2008

  8. [12]

    The anomalous threshold calculation in Schwarz (2008); Burridge & Taylor (2006) and Wilkinson (2017) is an application of Weissman’s spacing theorem (Weissman

    defines an anomalous threshold based on extreme value theory, a branch of probability theory that relates to the behaviour of extreme order statistics in a given sample (Galambos, Lechner & Simiu 2013). The anomalous threshold calculation in Schwarz (2008); Burridge & Taylor (2006) and Wilkinson (2017) is an application of Weissman’s spacing theorem (Weissman

    work page 2008

Show all 29 references
  1. [14]

    exhibits approximately the shape of an exponential distribution. Following Schwarz (2008), Burridge & Taylor (2006) and Wilkinson (2017), we start our anoma- lous threshold calculation from a subset of the points covering 50 per cent of them with the smallest anomalous scores ...

    2008

  2. [17]

    That is, for a small cluster to be a micro cluster, the number of data points in that cluster should be less than k

    One can interpret the value of k as the maximum permissible size for a micro cluster. That is, for a small cluster to be a micro cluster, the number of data points in that cluster should be less than k. Otherwise, the cluster is considered a typical cluster. (4) Figure 6 (d)- ...

    2019

  3. [18]

    Each scatterplot follows a negatively skewed distribution

    Figure 7 shows the pedestrian counts at 43 locations in the city of Melbourne at different times of the day. Each scatterplot follows a negatively skewed distribution. In general, weekdays display a bimodal distribution, while weekends follow a unimodal distribution. Now, the ...

    2005

  4. [19]

    Both days, 1 December 2018 and 31 December 2018, display an unusual rise later in the day

    Visual inspection also confirms the anomalous behaviour of these two scatterplots. Both days, 1 December 2018 and 31 December 2018, display an unusual rise later in the day. One selected day, 31 December 2018, is an obvious anomaly since it is the New Year’s Eve, and the assoc...

    2018

  5. [20]

    Anomalous days detected by the stray algorithm using scagnostics are marked in red colour. After detecting the anomalous scatterplots or the days with anomalous pedestrian behaviours, further investigation is carried out for each day to detect the locations with anomalous be- ...

    2019

  6. [22]

    It also can be used to identify anomalous time series within a large collection of streaming temporal data

    The anomalous time series detected by the stray algorithm using time series features are marked in red colour. It also can be used to identify anomalous time series within a large collection of streaming temporal data. Let W[t, t + w] represent a sliding window containing n nu...

    2015

  7. [23]

    In addition to a label, the stray algorithm also assigns an anomalous score to each data instance to indicate the degree of outlierness of each measurement

    In each plot, anomalies determined by the stray algorithm are represented in red colour demonstrate how the stray algorithm can assist in detecting anomalies present in other data structures using feature engineering. In addition to a label, the stray algorithm also assigns an...

    2017

  8. [209]

    Hodge, V & J Austin (2004)

    Wiley New York. Hodge, V & J Austin (2004). A survey of outlier detection methodologies. Artificial Intelligence Review 22(2), 85–126. Hofmann, H, H Wickham & K Kafadar (2017). Value plots: Boxplots for large data. Journal of Computational and Graphical Statistics 26(3), 469–4...

    2004 arXiv

  9. [283]

    A comparative evaluation of unsupervised anomaly detection algorithms for multivariate data

    Goldstein, M & S Uchida (2016). A comparative evaluation of unsupervised anomaly detection algorithms for multivariate data. PlOS ONE 11(4), e0152173. Grubbs, FE (1969). Procedures for detecting outlying observations in samples. Technometrics 11(1), 1–21. Gupta, M, J Gao, CC A...

    2016

  10. [1975]

    In the Leader algorithm, each cluster is a ball in the high-dimensional data space

    to form small clusters of points, prior to calculating nearest neighbour distance. In the Leader algorithm, each cluster is a ball in the high-dimensional data space. In the HDoutliers algorithm, the radius of this ball is selected such that it is well below the expected value...

    2019

  11. [1978]

    (Theorem 4.1) that is applicable to the distribution of data covered by the maximum domain of attraction of a Gumbel distribution. This requirement is satisfied by a wide range of distributions, ranging from those with light tails to moderately heavy tails that decrease to zer...

    2006

  12. [1993]

    Gao, J, W Hu, ZM Zhang, X Zhang & O Wu (2011)

    Springer US. Gao, J, W Hu, ZM Zhang, X Zhang & O Wu (2011). RKOF: robust kernel-based local outlier detection. In: Pacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, pp.270–

    2011

  13. [2002]

    This could be due to the parallelisability and memory access patterns of the two searching mechanisms

    have also reported a similar result for many algorithms based on kd-trees and many variants. This could be due to the parallelisability and memory access patterns of the two searching mechanisms. The brute force algorithm is easily parallelisable because it involves independen...

    2019

  14. [2005]

    Local parametric density-based outlier detection and ensemble learning with applications to malware detection

    IEEE, pp.157–164. Williams, KT (2016). “Local parametric density-based outlier detection and ensemble learning with applications to malware detection”. PhD thesis. The University of Texas at San Antonio. Zhang, R (2017). Performance of kd-tree vs brute-force nearest neighbor s...

    2016

  15. [2008]

    Some algorithms are application specific and take advantage of the underlying data structure or other domain-specific knowledge (Talagala et al

    and uncertain data (Aggarwal & Yu 2008). Some algorithms are application specific and take advantage of the underlying data structure or other domain-specific knowledge (Talagala et al. 2019b). More general algorithms without domain-specific knowledge are also available with t...

    2017

  16. [2014]

    Outlier detection

    will enable further insights into improved algorithm design. Anomaly detection problems commonly appear in many applications in different application domains. Therefore, it is hoped that different people with different knowledge levels will use the stray algorithm for many dif...

    2013

  17. [2015]

    and fraud detection (Talagala et al. 2019a). Ever-increasing computing resources and advanced data collection technologies that 2 Anomaly Detection in High Dimensional Data Talagala, Hyndman, Smith-Miles: 12 August 2019 3 emphasise real-time, large-scale data are other reasons...

    2019

  18. [2016]

    into high-dimensional data prior to the detection process for easy control. Under the high-dimensional data scenario, all attributes can be of the same data type or a mixture of different data types, such as categorical or numerical, which has a direct impact on the implementa...

    2017

  19. [2017]

    2014), machine learning and statistical domains (Hodge & Austin 2004), novelty detection (Pimentel et al

    or for specific data domains such as high-dimensional data, network data (Shahid, Naqvi & Qaisar 2015), temporal data (Gupta et al. 2014), machine learning and statistical domains (Hodge & Austin 2004), novelty detection (Pimentel et al. 2014), intrusion detection (Sabahi & Movaghar

    2015

  20. [2018]

    is available for the algorithm proposed by Wilkinson (2017). According to the R package implementation, the current version of the HDoutliers algorithm uses the next potential candidate for anomalies in calculating the anomalous threshold, in each iteration of the bottom-up se...

    2017

  21. [2019]

    Third, we demonstrate the wide applicability and usefulness of our stray algorithm, using various datasets

    package, stray (Talagala, Hyndman & Smith- Miles 2019), that implements the stray algorithm and related functions. Third, we demonstrate the wide applicability and usefulness of our stray algorithm, using various datasets. Anomaly Detection in High Dimensional Data Talagala, H...

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.