arxiv: 2604.22986 · v1 · submitted 2026-04-24 · 🌌 astro-ph.EP · astro-ph.IM

Recognition: unknown

A Machine Learning Approach to Meteor Classification

Samantha Hemmelgarn , Nicholas Moskovitz , Denis Vida

Authors on Pith no claims yet

Pith reviewed 2026-05-08 09:21 UTC · model grok-4.3

classification 🌌 astro-ph.EP astro-ph.IM

keywords meteor classificationmachine learningfactor analysisGaussian mixture modelmeteoroidshardness classificationcometary originasteroidal origin

0 comments

The pith

Machine learning on 28,000 meteor events produces a data-driven hardness classification for meteoroids.

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

This paper develops a machine learning framework to classify meteoroids using 13 observed parameters from the Global Meteor Network. By testing various normalization, dimensionality reduction, and clustering methods on over 28,000 events from the LO-CAMS network, it identifies factor analysis combined with a Gaussian mixture model as producing clusters that best match traditional physical models of meteoroid origin. The analysis reveals three underlying factors related to kinematics, activation thresholds, and size/geometry, with the activation factor best distinguishing asteroidal from cometary origins. It then introduces H_class, a hardness scheme ranging from dense iron meteoroids to soft cometary material, supported by an analytical formulation for future use. A sympathetic reader would care because it offers a scalable method to extract compositional information from large optical datasets without additional measurements.

Core claim

A combination of Factor Analysis (FA) and a Gaussian Mixture Model (GMM) applied to 13 directly observed meteor parameters from 28,177 events yields clusters most consistent with traditional models. Three FA-derived factors corresponding to meteoroid kinematics, activation thresholds, and size/geometry effects describe the underlying structure of meteoroid behavior. The activation factor emerges as the most discriminating for distinguishing asteroidal or cometary origin. Resulting 3-, 6-, and 11-cluster models reveal progressively finer compositional structure. From these, a physically motivated hardness classification scheme H_class is introduced as a data-driven extension of the Kb парамет

What carries the argument

Factor Analysis combined with a Gaussian Mixture Model applied to 13 observed parameters, which derives three factors and produces the H_class hardness scheme ranging from densest iron to softest cometary material.

If this is right

The activation threshold factor is the primary distinguisher between asteroidal and cometary meteoroids.
3-, 6-, and 11-cluster models reveal increasing detail from broad regimes to subdivisions within populations.
Application to nine well-studied meteor showers aids physical interpretation of the H_class groups in orbital space.
An analytical FA-GMM formulation enables direct application of the model to future datasets without retraining.
Machine learning methods can extract compositional information from modern optical meteor datasets at scale.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Future large-scale meteor surveys could apply this automated scheme to map distributions of solar system materials.
Cross-checking H_class assignments against spectroscopic data or recovered meteorites could validate or refine the groups.
Similar factor-analysis clustering might be tested on other optical transient datasets for compositional insights.
Links between H_class and orbital parameters could help trace meteoroid streams to specific parent bodies.

Load-bearing premise

That the three factors derived from factor analysis and the clusters from the Gaussian mixture model correspond to real physical differences in meteoroid composition and origin rather than artifacts of the chosen algorithms or normalization.

What would settle it

Independently determining the compositions of a set of meteors via spectroscopy or ground recovery of meteorites and checking whether their assigned H_class values match the expected hardness; systematic mismatch would show the clusters do not reflect physical reality.

Figures

Figures reproduced from arXiv: 2604.22986 by Denis Vida, Nicholas Moskovitz, Samantha Hemmelgarn.

**Figure 1.** Figure 1: LO-CAMS meteor detections from 2023 with the best 50% Median Fit Error (28,177 detections) color-coded based on their 𝐾𝑏 value calculated using Equation 1. These groupings are used to to establish a reference for evaluating the performance of our clustering methods. Beginning height is on the y-axis and initial velocity is on the x-axis. Blue points are cometary meteors where 𝐾𝑏 < 7.3, green points are car… view at source ↗

**Figure 2.** Figure 2: Histograms for each feature showing that the distributions in LO-CAMS 2023 data matches that of the entire GMN dataset. GMN data is in blue and LO-CAMS in orange. The GMN dataset was subject to the same median fit error cutoffs applied to the LO-CAMS dataset (Section 2.1) and contains 825,864 data points. The Y-axes show the normalized probability density for each feature, rather than raw counts. As a resu… view at source ↗

**Figure 3.** Figure 3: Heatmap of how 𝑆 changes as we increase the number of clusters for workflows using different dimensionality reduction techniques (FA, PCA, UMAP) and clustering methods (Agglomerative, BGMM, Birch, GMM, K-Means, and Spectral). The last row displays the mean silhouette score for each number of clusters and the last column shows the mean score for each clustering method. A requirement of any resulting algorit… view at source ↗

**Figure 4.** Figure 4: The separation between normalized 𝑆 and BIC (Δ) varies with cluster count, where noticeable Δs are found at 6 and 9-12 clusters. The combined behavior of these scores guided the selection of the number of clusters used in the GMM. Scores are normalized between the range [0, 1] to enable direct comparison since both metrics are defined on different scales. 𝑆 is shown in blue and BIC in orange. Annotations i… view at source ↗

**Figure 5.** Figure 5: 𝐻 𝑡beg vs. 𝑉init plots of rejected methods. These methods exhibited unrealistic behavior, such as diagonal or vertical separations and cluster overlap. We compared the groupings found by different combinations of dimensionality reduction techniques and clustering algorithms to the groupings revealed in this space by 𝐾𝑏 ( view at source ↗

**Figure 6.** Figure 6: HDBSCAN clustering results in 𝐻 𝑡beg vs. 𝑉init space for different combinations of dimensionality reduction techniques and min_cluster_size values. 𝑆 indicates modest clustering performance across all configurations. Panels a–b show FA results and panels c–d show PCA results. Cluster number −1 is assigned by the algorithm to noise points. All other cluster numbers represent definitive clusters identified b… view at source ↗

**Figure 7.** Figure 7: Magnitudes of features comprising the top five factor loadings for each of the kinematic, activation threshold, and size normalization factors. 4.1.1. Feature Dependencies and Their Impact on FA Several of the features we included are not independent. Specifically, 𝐸beg, 𝜌beg, 𝐿trail, and 𝑎decel are derived quantities that depend on other observable parameters. 𝐸beg is computed by first integrating atmosph… view at source ↗

**Figure 8.** Figure 8: Correlation matrix of the 13 features described in view at source ↗

**Figure 9.** Figure 9: Results from applying a GMM to FA reduced LO-CAMS data for (a) 3 clusters, (b) 6 clusters, and (c) 11 clusters. The left panel shows median activation scores (y-axis) vs. median kinematic scores (x-axis) for each cluster. The size of the scatter points scale with the size factor (smaller circle = smaller median size factor). The contours contain 66.7% (1𝜎) of the data points for each cluster. The right pan… view at source ↗

**Figure 10.** Figure 10: Sankey diagram of how the 𝐻class model grouped the populations of 9 meteor showers in the full GMN dataset from 2018 December 10 through 2025 May 16. To improve the clarity of this figure, only flow lines with a minimum of 250 meteors are shown. agrees with Ye and Vaubaillon (2022), who found that the brightest meteors observed during the 2022 TAH outburst were likely produced by centimeter-scale or large… view at source ↗

**Figure 11.** Figure 11: Evaluation of 𝐻class clusters in orbital space for the LO-CAMS 2023 dataset. The median Tisserand parameter (Levison, 1996) is plotted along the x-axis, where the annotations along the top of this axis indicate orbital classifications. The y-axis shows activation threshold scores, where annotations correspond to inferred material hardness (negative = Cometary, positive = Asteroidal). Colors denote cluster… view at source ↗

**Figure 12.** Figure 12: Physical interpretations of the 𝐻class model in factor space. Classes and their associated physical interpretations (Section 5.3) are shown in the legend on the right. Median activation score is on the y-axis, where it is inverted to maintain consistency with view at source ↗

**Figure 13.** Figure 13: Same as view at source ↗

read the original abstract

We use machine learning to develop a framework for classifying meteoroids based on 13 directly observed parameters from the Global Meteor Network. This method adds depth to the $K_{b}$ parameter, which uses only three parameters. We employ a semi-qualitative approach using 28,177 meteor events observed in 2023 by the Lowell Observatory Cameras for All-Sky Meteor Surveillance (LO-CAMS) network to evaluate multiple normalization, dimensionality-reduction, and clustering algorithms. We find that a combination of Factor Analysis (FA) and a Gaussian Mixture Model (GMM) results in clusters most consistent with traditional models. Three FA-derived factors corresponding to meteoroid kinematics, activation thresholds, and size/geometry effects describe the underlying structure of meteoroid behavior. The activation factor emerged as the most discriminating factor distinguishing whether a meteor is of asteroidal or cometary origin. Resulting 3, 6, and 11 cluster models reveal progressively finer compositional structure, from broad physical regimes to detailed subdivisions within cometary and asteroidal populations. From these results, we introduce a physically motivated hardness classification scheme: $H_{\mathrm{class}}$. $H_{\mathrm{class}}$ is a data-driven extension of $K_{b}$ which physically interprets clusters in terms of the densest iron meteoroids down to the softest cometary material. Application to nine well-studied meteor showers and analysis of clusters in orbital space aids in the physical interpretation of $H_{\mathrm{class}}$ groups. The $H_{\mathrm{class}}$ model is supported by an analytical FA-GMM formulation that enables application to future datasets. Our results demonstrate that machine learning methods can extract compositional information from modern optical meteor datasets at scale and offers a new framework for interpreting meteoroid populations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper runs factor analysis plus GMM on 13 meteor parameters to define a new H_class hardness scheme that refines Kb, but the physical claims rest on post-hoc labels without quantitative validation metrics.

read the letter

The main thing here is a practical pipeline that takes 28k meteor events from the LO-CAMS network, reduces 13 observed parameters via factor analysis, and clusters them with GMM to produce 3-, 6-, and 11-group models. From those they define H_class, which they present as a data-driven way to sort meteoroids from hard iron-like material down to soft cometary stuff, extending the older Kb parameter that used only three inputs. They also show the activation factor separates asteroidal from cometary origins and map the groups onto nine known showers plus orbital elements. That application to real events is the part that feels most concrete and potentially reusable for other large optical datasets. The analytical FA-GMM form they give for future data is a plus for reproducibility. The soft spots are exactly where the stress-test note flags them. The physical interpretations of the three factors and the clusters are assigned after the run, and the abstract only claims consistency with traditional models without showing agreement scores, confusion matrices, or any cross-check against independent composition data. The choice of three, six, or eleven clusters also looks driven by what produces interpretable groups rather than a formal criterion. Without those numbers or ablation results on normalization, it is hard to judge how much structure is physical versus an artifact of the 13 parameters or the algorithms. This is aimed at planetary scientists who process big meteor observation catalogs and want a scalable classification tool they can apply to new data. A reader who needs to organize similar datasets would get value from trying the pipeline and the H_class labels. It deserves a serious referee because the dataset size is respectable, the method is described clearly enough to reproduce, and the extension of Kb is a straightforward idea worth testing. Referees will almost certainly ask for the missing validation metrics, but the core work is coherent enough to review rather than desk-reject.

Referee Report

4 major / 3 minor

Summary. The manuscript applies Factor Analysis (FA) followed by Gaussian Mixture Model (GMM) clustering to 13 directly observed parameters from 28,177 meteors recorded by the LO-CAMS network in 2023. It identifies three latent factors (kinematics, activation thresholds, size/geometry), reports that the FA+GMM combination produces clusters most consistent with traditional models, and introduces H_class as a data-driven hardness classification extending the three-parameter Kb scheme. The work includes qualitative support via nine meteor showers and orbital-element analysis, plus an analytical FA-GMM formulation for future datasets.

Significance. If the unsupervised clusters demonstrably recover genuine compositional and origin distinctions rather than statistical artifacts, the framework would provide a scalable, reproducible method for interpreting large optical meteor datasets beyond the limitations of Kb. The explicit analytical formulation is a positive feature for reproducibility and extension to new observations.

major comments (4)

[Abstract] Abstract: the central claim that FA+GMM clusters are 'most consistent with traditional models' is unsupported by any quantitative metric (agreement score, confusion matrix, or statistical test against Kb or other classifications). This absence directly undermines the physical motivation asserted for the H_class scheme.
[Results] The selection of exactly 3, 6, and 11 clusters is presented without reported validation criteria (silhouette scores, BIC, elbow plots, or stability across random seeds). Because H_class is defined from these clusters, the lack of justification makes the number of groups appear post-hoc and weakens the claim of progressively finer compositional structure.
[Methods / Results] The assignment of physical labels to the three FA factors (kinematics, activation thresholds, size/geometry) and the interpretation that the activation factor distinguishes asteroidal vs. cometary origin occur after the FA step. No a priori physical model or independent compositional labels are used to validate these interpretations, leaving open the possibility that the factors reflect parameter correlations or normalization choices instead of ablation physics.
[Methods] No ablation or sensitivity tests are described for the choice among normalization schemes, the impact of the specific 13 parameters, or robustness to subsetting the 28,177-event sample. Given that the method is positioned as adding depth to Kb, such tests are required to establish that the recovered structure is not an artifact of preprocessing.

minor comments (3)

[Abstract / Methods] The term 'semi-qualitative approach' is used in the abstract but not defined in the methods; a brief clarification of what this entails (e.g., which steps are quantitative vs. interpretive) would improve clarity.
[Results] The manuscript would benefit from explicit reporting of cluster-assignment uncertainties or posterior probabilities from the GMM, especially when mapping clusters to H_class labels.
[Discussion] A short comparison table or figure overlaying H_class against Kb values for the nine meteor showers would make the claimed consistency more transparent.

Simulated Author's Rebuttal

4 responses · 0 unresolved

Thank you for the referee's thoughtful and constructive comments, which help improve the clarity and rigor of our work. We respond to each major comment point by point below, indicating revisions where appropriate.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that FA+GMM clusters are 'most consistent with traditional models' is unsupported by any quantitative metric (agreement score, confusion matrix, or statistical test against Kb or other classifications). This absence directly undermines the physical motivation asserted for the H_class scheme.

Authors: We thank the referee for highlighting this. While the manuscript supports the claim through qualitative analysis of meteor showers and orbital elements, we agree a quantitative metric would strengthen it. In the revised manuscript, we will include an agreement score or confusion matrix comparing H_class to Kb classifications, along with a statistical test to quantify consistency. revision: yes
Referee: [Results] The selection of exactly 3, 6, and 11 clusters is presented without reported validation criteria (silhouette scores, BIC, elbow plots, or stability across random seeds). Because H_class is defined from these clusters, the lack of justification makes the number of groups appear post-hoc and weakens the claim of progressively finer compositional structure.

Authors: The cluster numbers were chosen to progressively refine the traditional three-class Kb model (3 clusters), with 6 and 11 providing finer divisions based on observed data structure. We will revise to report validation criteria including silhouette scores, BIC, and cluster stability across seeds to justify these choices rigorously. revision: yes
Referee: [Methods / Results] The assignment of physical labels to the three FA factors (kinematics, activation thresholds, size/geometry) and the interpretation that the activation factor distinguishes asteroidal vs. cometary origin occur after the FA step. No a priori physical model or independent compositional labels are used to validate these interpretations, leaving open the possibility that the factors reflect parameter correlations or normalization choices instead of ablation physics.

Authors: The factor interpretations are derived from the parameter loadings and aligned with established meteoroid physics literature on kinematics, ablation thresholds, and size effects. As an unsupervised approach, labels are necessarily post-hoc. We will expand the discussion in the revision to include more explicit references to physical models and note the exploratory nature, while acknowledging potential influences from correlations. revision: partial
Referee: [Methods] No ablation or sensitivity tests are described for the choice among normalization schemes, the impact of the specific 13 parameters, or robustness to subsetting the 28,177-event sample. Given that the method is positioned as adding depth to Kb, such tests are required to establish that the recovered structure is not an artifact of preprocessing.

Authors: We agree on the importance of these tests. The original analysis considered multiple algorithms, but we will add sensitivity analyses in the revision, including variations in normalization, parameter importance via ablation studies, and robustness checks on data subsets to demonstrate that the recovered factors and clusters are stable and not artifacts. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper applies unsupervised FA and GMM clustering directly to 13 observed parameters from the 2023 LO-CAMS dataset. Factors (kinematics, activation thresholds, size/geometry) and the 3/6/11-cluster models emerge from the data without any presupposed physical labels or H_class scheme. H_class is defined afterward as a post-hoc interpretive extension of the existing Kb parameter, with qualitative consistency checks against meteor showers and orbital elements. No load-bearing step reduces by construction to a fitted constant, self-citation chain, or ansatz smuggled from prior work; the method is self-contained against the input observations.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that the 13 observed parameters encode compositional information and that unsupervised clusters map to physical regimes; free parameters include the number of clusters and algorithm choices.

free parameters (1)

Number of clusters
Authors select 3, 6, and 11 clusters to show progressive structure; these values are chosen rather than derived from a uniqueness criterion.

axioms (1)

domain assumption The 13 directly observed parameters from the Global Meteor Network capture the relevant physical properties of meteoroids.
Invoked when applying dimensionality reduction and clustering to classify origin and hardness.

invented entities (1)

H_class no independent evidence
purpose: Data-driven hardness classification extending Kb
New scheme defined from the FA-GMM clusters; no independent falsifiable prediction supplied in the abstract.

pith-pipeline@v0.9.0 · 5615 in / 1391 out tokens · 24870 ms · 2026-05-08T09:21:24.388397+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

56 extracted references · 52 canonical work pages · 1 internal anchor

[1]

Astronomy & Astrophysics 696, A69

Ashimbekova,A.,Vaubaillon,J.,Koten,P.,2025. Towardsadefinitionofameteorcluster:Detectionofmeteorclustersfrommeteororbitdatabases. A&A 696, A69. doi:10.1051/0004-6361/202452445,arXiv:2503.16157. Babadzhanov, P.B.,

work page doi:10.1051/0004-6361/202452445 2025
[2]

A&A 384, 317–321

Fragmentation and densities of meteoroids. A&A 384, 317–321. doi:10.1051/0004-6361:20020010. Babadzhanov, P.B., Williams, I.P., Kokhirova, G.I.,

work page doi:10.1051/0004-6361:20020010
[3]

Cognitive Science36(5), 757–798 (2012).https://doi.org/10.1111/j

Meteor showers associated with 2003EH1. MNRAS 386, 2271–2277. doi:10.1111/j. 1365-2966.2008.13202.x. Barber, D.,

work page doi:10.1111/j 2008
[4]

A&A 473, 661–672

Atmospheric deceleration and light curves of Draconid meteors and implications for the structure of cometary dust. A&A 473, 661–672. doi:10.1051/0004-6361:20078131. Borovička, J., Spurný, P., Shrbený, L.,

work page doi:10.1051/0004-6361:20078131
[5]

Data on 824 fireballs observed by the digital cameras of the European Fireball Network in 2017-

2017
[6]

Analysis of orbital and physical properties of centimeter-sized meteoroids

II. Analysis of orbital and physical properties of centimeter-sized meteoroids. A&A 667, A158. doi:10.1051/0004-6361/202244197, arXiv:2209.11254. Buccongello, N., Brown, P.G., Vida, D., Pinhas, A.,

work page doi:10.1051/0004-6361/202244197
[7]

Icarus 410, 115907

A physical survey of meteoroid streams: Comparing cometary reservoirs. Icarus 410, 115907. doi:10.1016/j.icarus.2023.115907,arXiv:2312.00897. Hemmelgarn et al.:Preprint submitted to ElsevierPage 26 of 35 Machine Learning Meteor Classification Campello, R., Moulavi, D., Sander, J.,

work page doi:10.1016/j.icarus.2023.115907 2023
[8]

Fireball end heights: A diagnostic for the structure of meteoric material. J. Geophys. Res. 81, 6257–6275. doi:10.1029/JB081i035p06257. Cherkassky, V., Ma, Y.,

work page doi:10.1029/jb081i035p06257
[9]

Neural Computation 15, 1691–1714

Comparison of model selection for regression. Neural Computation 15, 1691–1714. URL:https://doi.org/10.1162/089976603321891864, doi:10.1162/089976603321891864, arXiv:https://direct.mit.edu/neco/article-pdf/15/7/1691/815658/089976603321891864.pdf. Colas, F., Zanda, B., Bouley, S., Jeanne, S., Malgoyre, A., Birlan, M., Blanpain, C., Gattacceca, J., Jorda, L...

work page doi:10.1162/089976603321891864
[10]

Astronomy and Astrophysics , keywords =

FRIPON: a worldwide network to track incoming meteoroids. A&A 644, A53. doi:10.1051/0004-6361/202038649,arXiv:2012.00616. Cordonnier, L.E., Obenberger, K.S., Holmes, J.M., Taylor, G.B., Vida, D.,

work page doi:10.1051/0004-6361/202038649 2012
[11]

Journal of Geophysical Research: Space Physics 129, e2024JA032643

Not so fast: A new catalog of meteor persistent trains. Journal of Geophysical Research: Space Physics 129, e2024JA032643. URL:https:// agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2024JA032643, doi:https://doi.org/10.1029/2024JA032643, arXiv:https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2024JA032643. e2024JA032643 2024JA032643. Dempster, A.P...

work page doi:10.1029/2024ja032643
[12]

Journal of the Royal Statistical Society: Series B (Methodological) , author =

Maximum likelihood from incomplete data via the em algorithm. Journal of the Royal Statistical Society: Series B (Methodological) 39, 1–22. URL:https://doi.org/10.1111/j.2517-6161.1977.tb01600.x, doi:10. 1111/j.2517-6161.1977.tb01600.x. Devillepoix, H., Cupák, M., Bland, P., Sansom, E., Towner, M., Howie, R., Hartig, B., Jansen-Sturgeon, T., Shober, P., A...

work page doi:10.1111/j.2517-6161.1977.tb01600.x 1977
[13]

Icarus 406, 115718

A machine learning classification of meteorite spectra applied to understanding asteroids. Icarus 406, 115718. URL:https://www.sciencedirect.com/science/article/pii/S0019103523002956, doi:https: //doi.org/10.1016/j.icarus.2023.115718. Egal, A., Brown, P.G., Rendtel, J., Campbell-Brown, M., Wiegert, P.,

work page doi:10.1016/j.icarus.2023.115718 2023
[14]

A&A 640, A58

Activity of the Eta-Aquariid and Orionid meteor showers. A&A 640, A58. doi:10.1051/0004-6361/202038115,arXiv:2006.08576. Egal,A.,Wiegert,P.,Brown,P.,Moser,D.,Campbell-Brown,M.,Moorhead,A.,Ehlert,S.,Moticska,N.,2019. Meteorshowermodeling:Pastand futuredraconidoutbursts. Icarus330,123–141. URL:https://www.sciencedirect.com/science/article/pii/S0019103518308...

work page doi:10.1051/0004-6361/202038115 2006
[15]

Monthly Notices of the Royal Astronomical Society 469, S39–S44

Fractal dust constrains the collisional history of comets. Monthly Notices of the Royal Astronomical Society 469, S39–S44. URL:https://doi.org/10.1093/mnras/stx971, doi:10.1093/mnras/stx971, arXiv:https://academic.oup.com/mnras/article-pdf/469/Suppl_2/S39/17484368/stx971.pdf. Gural, P.S.,

work page doi:10.1093/mnras/stx971
[16]

R., Millman, K

Array programming with NumPy. Nature 585, 357–362. URL: Hemmelgarn et al.:Preprint submitted to ElsevierPage 27 of 35 Machine Learning Meteor Classification https://doi.org/10.1038/s41586-020-2649-2, doi:10.1038/s41586-020-2649-2. Hemmelgarn, S., Moskovitz, N., Pilorz, S., Jenniskens, P.,

work page doi:10.1038/s41586-020-2649-2
[17]

Hunter, J.D.,

doi:10.3847/PSJ/ad8346. Hunter, J.D.,

work page doi:10.3847/psj/ad8346
[18]

Computing in Science and Engineering , keywords =

Matplotlib: A 2d graphics environment. Computing in Science & Engineering 9, 90–95. doi:10.1109/MCSE.2007.55. Jenniskens,P.,2004. 2003eh1isthequadrantidshowerparentcomet. TheAstronomicalJournal127,3018. URL:https://doi.org/10.1086/ 383213, doi:10.1086/383213. Jenniskens, P.,

work page doi:10.1109/mcse.2007.55 2007
[19]

Icarus 216, 40–61

Cams: Cameras for allsky meteor surveillance to establish minor meteor showers. Icarus 216, 40–61. URL:https://www.sciencedirect.com/science/article/pii/ S0019103511003290, doi:https://doi.org/10.1016/j.icarus.2011.08.012. Jewitt, D., Li, J.,

work page doi:10.1016/j.icarus.2011.08.012 2011
[20]

AJ 140, 1519–1527

Activity in Geminid Parent (3200) Phaethon. AJ 140, 1519–1527. doi:10.1088/0004-6256/140/5/1519, arXiv:1009.2710. Jo, H., Ishiguro, M.,

work page doi:10.1088/0004-6256/140/5/1519
[21]

A&A 683, A68

Dynamical study of Geminid formation assuming a rotational instability scenario. A&A 683, A68. doi:10.1051/ 0004-6361/202347898,arXiv:2401.03682. Jolliffe, I.,

work page arXiv
[22]

Aharonian, A

The parent bodies of the Quadrantid meteoroid stream. A&A 470, 1123–1136. doi:10.1051/0004-6361: 20077329. Kennedy, A.B.W., Sankey, H.R.,

work page doi:10.1051/0004-6361:
[23]

Minutes of the Proceedings of the Institu- tion of Civil Engineers 134, 278–312

The thermal efficiency of steam engines. Minutes of the Proceedings of the Institu- tion of Civil Engineers 134, 278–312. URL:https://doi.org/10.1680/imotp.1898.19100, doi:10.1680/imotp.1898.19100, arXiv:https://www.emerald.com/jmipi/article-pdf/134/1898/278/2558573/imotp_1898_19100.pdf. Kikwaya, J.B., Campbell-Brown, M., Brown, P.G.,

work page doi:10.1680/imotp.1898.19100
[24]

Perego, A

Bulk density of small meteoroids. A&A 530, A113. doi:10.1051/0004-6361/ 201116431. Koten,P.,Borovička,J.,Spurný,P.,Betlem,H.,Evans,S.,2004. Atmospherictrajectoriesandlightcurvesofshowermeteors. A&A428,683–690. doi:10.1051/0004-6361:20041485. Levison, H.F.,

work page doi:10.1051/0004-6361/ 2004
[25]

Nature Astronomy 8, 60–68

Thermal decomposition as the activity driver of near-Earth asteroid (3200) Phaethon. Nature Astronomy 8, 60–68. doi:10.1038/s41550-023-02091-w,arXiv:2207.08968. McInnes, L., Healy, J., Melville, J.,

work page doi:10.1038/s41550-023-02091-w
[26]

UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

Umap: Uniform manifold approximation and projection for dimension reduction. URL:https: //arxiv.org/abs/1802.03426,arXiv:1802.03426. Mcintosh, B.A.,

work page internal anchor Pith review arXiv
[27]

Icarus 86, 299–304

Comet p/machholz and the quadrantid meteor stream. Icarus 86, 299–304. URL:https://www.sciencedirect.com/ science/article/pii/001910359090219Y, doi:https://doi.org/10.1016/0019-1035(90)90219-Y. McIntosh, B.A., Hajduk, A.,

work page doi:10.1016/0019-1035(90)90219-y
[28]

MNRAS 205, 931–943

Comet Halley meteor stream: a new model. MNRAS 205, 931–943. doi:10.1093/mnras/205.4.931. McIntosh, B.A., Jones, J.,

work page doi:10.1093/mnras/205.4.931
[29]

MNRAS 235, 673–693

The Halley comet meteor stream - Numerical modelling of its dynamic evolution. MNRAS 235, 673–693. doi:10.1093/mnras/235.3.673. McKinney, W.,

work page doi:10.1093/mnras/235.3.673
[30]

2010 , keywords =

doi:10.25080/Majora-92bf1922-00a. Miller, G.J., Brandler, S., Roman, C.P., Yang, K.,

work page doi:10.25080/majora-92bf1922-00a
[31]

CRC press

Handbook of research methods in public administration. CRC press. Mills,T.,Brown,P.G.,Mazur,M.J.,Vida,D.,Gural,P.S.,Moorhead,A.V.,2021.Ironrain:measuringtheoccurrencerateandoriginofsmallironme- teoroidsatearth. MonthlyNoticesoftheRoyalAstronomicalSociety508,3684–3696. URL:https://doi.org/10.1093/mnras/stab2743, doi:10.1093/mnras/stab2743,arXiv:https://aca...

work page doi:10.1093/mnras/stab2743 2021
[32]

(Ed.), Meteoroids 2001 Conference, pp

The AKM video meteor network, in: Warmbein, B. (Ed.), Meteoroids 2001 Conference, pp. 315–318. Molau, S., Barentsen, G.,

2001
[33]

(Eds.), Meteoroids 2013, pp

Status and history of the IMO Video Meteor Network, in: Jopek, T.J., Rietmeijer, F.J.M., Watanabe, J., Williams, I.P. (Eds.), Meteoroids 2013, pp. 297–305. Neslušan,L.,Hajduková,M.,Jakubík,M.,2013a. Meteor-showercomplexofasteroid2003EH1comparedwiththatofcomet96P/Machholz. A&A 560, A47. doi:10.1051/0004-6361/201322228. Neslušan, L., Kaňuchová, Z., Tomko, D...

work page doi:10.1051/0004-6361/201322228 2013
[34]

arXiv e-prints , arXiv:2507.01501doi:10.48550/arXiv.2507.01501,arXiv:2507.01501

Meteoroid stream identification with HDBSCAN unsupervised clustering algorithm. arXiv e-prints , arXiv:2507.01501doi:10.48550/arXiv.2507.01501,arXiv:2507.01501. Peña-Asensio, E., Trigo-Rodríguez, J.M., Grèbol-Tomàs, P., Regordosa-Avellana, D., Rimola, A.,

work page doi:10.48550/arxiv.2507.01501
[35]

Nicolás, The bar derived category of a curved dg algebra, Journal of Pure and Applied Algebra 212 (2008) 2633–2659

Deep machine learning for meteor monitoring:Advanceswithtransferlearningandgradient-weightedclassactivationmapping. Planet.SpaceSci.238,105802. doi:10.1016/j. pss.2023.105802,arXiv:2310.16826. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., ...

work page doi:10.1016/j 2023
[36]

1987 , issue_date =

Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20, 53–65. URL:https://www.sciencedirect.com/science/article/pii/0377042787901257, doi:https: //doi.org/10.1016/0377-0427(87)90125-7. Hemmelgarn et al.:Preprint submitted to ElsevierPage 28 of 35 Machine Learning Meteor C...

work page doi:10.1016/0377-0427(87)90125-7
[37]

Monthly Notices of the Royal Astronomical Society 456, 78–84

A preliminary numerical model of the geminid meteoroid stream. Monthly Notices of the Royal Astronomical Society 456, 78–84. URL:https://doi.org/10.1093/mnras/stv2626, doi:10.1093/mnras/stv2626, arXiv:https://academic.oup.com/mnras/article-pdf/456/1/78/3658524/stv2626.pdf. Sicking, W.,

work page doi:10.1093/mnras/stv2626
[38]

WGN, Journal of the International Meteor Organization 37, 55–62

A meteor shower catalog based on video observations in 2007-2008. WGN, Journal of the International Meteor Organization 37, 55–62. SonotaCo, Masuzawa, T., Sekiguchi, T., Miyoshi, T., Fujiwara, Y., Maeda, K., Uehara, S.,

2007
[39]

general intelligence,

"general intelligence," objectively determined and measured. The American Journal of Psychology 15, 201–292. URL: http://www.jstor.org/stable/1412107. Sugar, G., Moorhead, A., Brown, P., Cooke, W.,

work page arXiv
[40]

MAPS 52, 1048–1059

Meteor shower detection with density-based clustering. MAPS 52, 1048–1059. doi:10.1111/maps.12856,arXiv:1702.02656. The Pandas Development Team,

work page doi:10.1111/maps.12856
[41]

pandas development team, pandas-dev/pandas: Pandas (Feb

pandas-dev/pandas: Pandas. URL:https://doi.org/10.5281/zenodo.3509134, doi:10.5281/ zenodo.3509134. Thurstone, L.,

work page doi:10.5281/zenodo.3509134
[42]

Vaubaillon, J., Colas, F., Jorda, L.,

A method for synthesis of factor analysis studies doi:https://doi.org/10.21236/AD0047524. Vaubaillon, J., Colas, F., Jorda, L.,

work page doi:10.21236/ad0047524
[43]

A new method to predict meteor showers. I. Description of the model. A&A 439, 751–760. doi:10.1051/0004-6361:20041544. Čapek, D., Koten, P., Spurný, P., Shrbený, L.,

work page doi:10.1051/0004-6361:20041544
[44]

Astronomy & Astrophysics 666, A144

Ejection velocities, age, and formation process of SPE meteoroid cluster. A&A 666, A144. doi:10.1051/0004-6361/202243055,arXiv:2207.14029. Verniani, F.,

work page doi:10.1051/0004-6361/202243055
[45]

Monthly Notices of the Royal Astronomical So- ciety 515, 2322–2339

Comput- ing optical meteor flux using global meteor network data. Monthly Notices of the Royal Astronomical So- ciety 515, 2322–2339. URL:https://doi.org/10.1093/mnras/stac1766, doi:10.1093/mnras/stac1766, arXiv:https://academic.oup.com/mnras/article-pdf/515/2/2322/45192083/stac1766.pdf. Vida, D., Brown, P.G., Campbell-Brown, M., Egal, A.,

work page doi:10.1093/mnras/stac1766
[46]

Icarus 408, 115842

First holistic modelling of meteoroid ablation and fragmentation: A case study of the Orionids recorded by the Canadian Automated Meteor Observatory. Icarus 408, 115842. doi:10.1016/j.icarus.2023.115842, arXiv:2310.12776. Vida,D.,Gural,P.S.,Brown,P.G.,Campbell-Brown,M.,Wiegert,P.,2019.Estimatingtrajectoriesofmeteors:anobservationalmontecarloapproach – i. ...

work page doi:10.1016/j.icarus.2023.115842 2023
[47]

, keywords =

The Global Meteor Network – Methodology and first results. Monthly Notices of the Royal Astronomical Society 506, 5046–5074. URL:https://doi.org/10.1093/mnras/stab2008, doi:10.1093/mnras/stab2008, arXiv:https://academic.oup.com/mnras/article-pdf/506/4/5046/39627793/stab2008.pdf. Vinh, N.X., Epps, J., Bailey, J.,

work page doi:10.1093/mnras/stab2008
[48]

Information theoretic measures for clusterings comparison: is a correction for chance necessary?, in: Proceedings of the 26th Annual International Conference on Machine Learning, Association for Computing Machinery, New York, NY, USA. p. 1073–1080. URL:https://doi.org/10.1145/1553374.1553511, doi:10.1145/1553374.1553511. Vojáček,V.,Borovička,J.,Koten,P.,S...

work page doi:10.1145/1553374.1553511 2015
[49]

Icarus 441, 116698

A statistical approach to quantifying uncertainty in meteoroid physical properties. Icarus 441, 116698. doi:10.1016/j.icarus.2025.116698,arXiv:2506.21701. Waskom, M.L.,

work page doi:10.1016/j.icarus.2025.116698 2025
[50]

Waskom , title =

URL:https://doi.org/10.21105/ joss.03021, doi:10.21105/joss.03021. Whipple, F.L.,

work page doi:10.21105/joss.03021
[51]

Williams, I.P., Wu, Z.,

doi:10.1086/145416. Williams, I.P., Wu, Z.,

work page doi:10.1086/145416
[52]

Monthly Notices of the Royal Astronomical Society 262, 231–248

The geminid meteor stream and asteroid 3200 phaethon. Monthly Notices of the Royal Astronomical Society 262, 231–248. URL:https://doi.org/10.1093/mnras/262.1.231, doi:10.1093/mnras/262.1.231, arXiv:https://academic.oup.com/mnras/article-pdf/262/1/231/18539577/mnras262-0231.pdf. Ye, Q., Vaubaillon, J.,

work page doi:10.1093/mnras/262.1.231
[53]

The 2022 encounter of the outburst material from comet 73p/schwassmann–wachmann

2022
[54]

URL:https://doi.org/10.1093/mnrasl/slac070,doi:10.1093/mnrasl/slac070, arXiv:https://academic.oup.com/mnrasl/article-pdf/515/1/L45/54645478/slac070.pdf

Monthly Notices of the RoyalAstronomicalSociety:Letters515,L45–L49. URL:https://doi.org/10.1093/mnrasl/slac070,doi:10.1093/mnrasl/slac070, arXiv:https://academic.oup.com/mnrasl/article-pdf/515/1/L45/54645478/slac070.pdf. Zou, H., Hastie, T., Tibshirani, R.,

work page doi:10.1093/mnrasl/slac070
[55]

1214/009053607000000127, doi:10.1214/009053607000000127

URL:http://dx.doi.org/10. 1214/009053607000000127, doi:10.1214/009053607000000127. Zubović,D.,Vida,D.,Gural,P.,Šegon,D.,2015. Advancesinthedevelopmentofalow-costvideometeorstation. ProceedingsoftheInternational Meteor Conference, Mistelbach, Austria , 27–30. Hemmelgarn et al.:Preprint submitted to ElsevierPage 29 of 35 Machine Learning Meteor Classificati...

work page doi:10.1214/009053607000000127 2015
[56]

(15) The factor loadings𝑊and feature-specific noise variancesΨare estimated from the training data by maximizing the FA likelihood, after which the Varimax rotation is applied to the loadings. This rotation is an input parameter to theFactorAnalysisfunction and is incorporated directly into the factor loadings𝑊, so we will not go through the mathematical ...

work page arXiv 1958