Pith. sign in

REVIEW 4 major objections 5 minor 61 references

The Geography of Transportation Cybersecurity: Visitor Flows, Industry Clusters, and Spatial Dynamics

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The paper claims that US transportation cybersecurity visitor flows will rise 14.16% on average, and that geolocation plus education, not other social factors, most strongly shape where these industries cluster.

desk verdict Useful descriptive geography, but the headline forecast and factor-importance claims are circular and the evaluation metrics are internally inconsistent. read the letter →

arxiv 2505.08822 v1 pith:VPGRXUOZ submitted 2025-05-12 cs.CY cs.LGphysics.soc-ph

classification cs.CYcs.LGphysics.soc-ph
keywords transportationcybersecurityvisitorflowsspatialclusteringindustryclustersgraphconvolutionalnetworkTransformerGeoShapleysocioeconomicfactors
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 the three industries forming the transportation cybersecurity ecosystem (cybersecurity, automotive, and transportation and logistics) are geographically organized into distinct, measurable clusters, and that where people travel to visit these businesses can be predicted. If true, it would give planners a data-driven way to see where transportation-cybersecurity jobs and activity are heading, and which regional conditions attract them. The paper's central empirical claims are that geolocation and education are the strongest factors shaping cluster formation, and that overall US visitor flow in these industries will rise about 14.16% in the next period. The authors argue this makes the geography of these industries something that can be tracked and anticipated rather than only described after the fact.

What carries the argument

The central mechanism is BiTransGCN, a hybrid deep-learning model that first passes weekly visitor-flow counts through a graph convolutional network to capture spatial structure among regions, then through an attention-based Transformer to capture long-range temporal dependencies, and finally maps the combined representation to next-period visitor counts. The attribution machinery is GeoShapley, a spatial version of Shapley values that treats geographic location as a player in a coalition and can therefore separate the intrinsic effect of place from the effects of education, housing, crime, work, health, and economy. Spatial clustering is measured with K-means on normalized visitor-flow levels and global and local Moran's I statistics, with gradient-boosted trees as the regression model on which GeoShapley is computed.

What would settle it

Obtain an independent year of visitor-flow or establishment-level employment data for the same industries and re-run the forecast and the factor-importance analysis; the central claims would be overturned if the next-period growth rates do not materialize or if geolocation and education are not the top factors in that independent data.

Watch

Extended reading notes

Core claim

The paper claims that visitor flows in the transportation cybersecurity ecosystem follow distinct, industry-specific spatial patterns: automotive visits are the most voluminous and long-distance, centered on a southern corridor anchored by Texas; cybersecurity visits are more localized, with isolated hotspots in states such as Oregon and Colorado; and transportation and logistics visits are the most evenly spread, with the Midwest as a core region. It further claims that, when all social factors are considered together, geolocation and education dominate cluster formation, while the influence of health, housing, crime, work, and economy varies by industry. Finally, the BiTransGCN forecast claims US visitor flow across these industries will grow by an average of 14.16% in the next term, with automotive at 16.72%, cybersecurity highly volatile at 58.14% on average while some states decline, and transportation and logistics declining by 18.77%.

Load-bearing premise

The whole empirical chain assumes that smartphone-based visitor counts, grouped into the three industries through broad industry-classification codes, faithfully represent actual business visitor flows in the US.

Editorial extensions

If this is right

  • If the forecast is correct, transportation-cybersecurity activity will grow in all 51 states in the next term, with an average increase of 14.16%.
  • The forecast implies a structural shift within the ecosystem: automotive and cybersecurity visits grow while transportation and logistics visits decline, on average, by 18.77%.
  • Texas should consolidate its position as a leading hub, since it ranks highest in visitor flow and in clustering levels across multiple sectors.
  • Because geolocation and education dominate cluster formation, regional workforce and higher-education policy become natural levers for attracting these industries.
  • The combination of spatial clustering analysis and deep-learning prediction gives a method for anticipating, rather than only reacting to, regional industry shifts in cybersecurity-adjacent sectors.

Reading between the lines

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

  • A testable extension the authors do not run: use changes in regional bachelor's-degree attainment as a leading indicator for future TCI cluster growth; if education is genuinely causal, attainment changes should precede cluster shifts.
  • The forecast likely inherits biases from smartphone-location panels, so an independent check against payroll, employment, or establishment-level data would be a natural next test of the central claims.
  • The same pipeline could be reapplied to other cyber-adjacent sectors, or to the same three sectors in other countries, wherever cell-phone-based origin-destination data are available.
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

4 major / 5 minor

Summary. The paper examines 2022 SafeGraph weekly visitor-flow data for three transportation-cybersecurity-related industries (cybersecurity, automotive, transportation and logistics), develops a BiTransGCN deep-learning model to forecast state-level visitor flows, applies K-means and Moran's I for spatial clustering analysis, and uses XGBoost with GeoShapley to rank socioeconomic factors. The headline claims are a 14.16% average increase in US visitor flows in the next period and that geolocation and education are the most significant factors influencing industry cluster formation.

Significance. The research question is timely and the data integration is ambitious: linking visitor flows, industrial clustering, and socioeconomic conditions could inform transportation-cybersecurity workforce and infrastructure planning. The paper also attempts a methodological bridge between deep-learning forecasting and spatial explainability. However, as written, the central empirical claims are not supported. The factor-importance analysis is circular because it explains the model's own predictions rather than observed cluster formation, the reported factor rankings are internally inconsistent, and the forecast is presented without uncertainty quantification or baseline comparison. These issues are load-bearing, so the paper's contribution would require substantial reanalysis and reframing before it can be accepted.

major comments (4)
  1. [Sec. 5.4, Eq. (12), Fig. 13] The factor-importance claim is circular. Section 5.4 states that the dependent variable is "the previously predicted rates of change in visitation volumes for each industry," and Eq. (12) decomposes the model's predicted y-hat. Consequently, the GeoShapley results describe what the fitted BiTransGCN forecaster is sensitive to, not what actually explains observed industry cluster formation or observed flow changes. The abstract and conclusion nonetheless claim that geolocation and education are the most significant factors influencing industry cluster formation. To support that claim, the analysis must use observed, not predicted, outcomes as the dependent variable, or the claims must be explicitly reframed as an interpretability analysis of the model.
  2. [Abstract, Sec. 5.4, Conclusion] The reported factor rankings are internally inconsistent. The abstract and conclusion state that geolocation and education are the most significant factors, but Sec. 5.4 says "geolocation and work-related factors are the most significant driver," and Fig. 13 ranks work above education for the overall TCI analysis. For the transportation and logistics sector, education is reported as most influential; for cybersecurity, housing is second. The paper cannot present these conflicting results without reconciliation. This inconsistency undermines the headline claim.
  3. [Sec. 5.3, Table 4, Fig. 11] The 14.16% average increase in visitor flow is not adequately supported. The model is used to forecast "the next week" with a 4:1 temporal split, but the text then interprets week-on-week changes of predicted values as a long-term projection; the forecast horizon is undefined. No confidence intervals, no baseline comparison (e.g., historical average or ARIMA), and no out-of-sample evaluation across multiple horizons are provided. The cybersecurity sector has MAE 0.506 and MAPE 24.40%, and some states are reported to grow by over 500%, which suggests instability. The headline forecast requires a defined horizon, uncertainty estimates, and baseline comparisons.
  4. [Table 1, Sec. 3.1] The validity of the industry categories is load-bearing but not established. The cybersecurity category includes NAICS codes such as 561622 (Locksmiths) and 541690 (Other Scientific and Technical Consulting Services), which are broad and may not represent cybersecurity activity. SafeGraph smartphone-location panels are known to undersample certain demographics. Because the central claims depend on these visitor-flow measures, the paper should report robustness checks (e.g., excluding ambiguous codes, sensitivity to panel composition) or clearly discuss the limitations of the mapping.
minor comments (5)
  1. [Sec. 3.2] The description of K-means says "an optimal number of K=6 clusters" but no method for selecting K is given; the cluster-level bin boundaries in Sec. 5.2 also appear arbitrary. Please justify these choices.
  2. [Sec. 5.2] There is a figure-numbering inconsistency: the text refers to "Fig. 9(d-f)" when discussing maps that are labeled as part of Fig. 8, and the bivariate global Moran's I maps are described as "global and local" without a clear distinction. Please correct the cross-references and clarify the Moran's I interpretation.
  3. [Fig. 1 caption] The caption contains a typo: "CGB density maps" should be "CBG density maps." In addition, Table 2 has a typo in the header "F actor."
  4. [Sec. 4.1] The term "bidirectional" in BiTransGCN is not defined. The architecture description presents a standard Transformer with multi-head attention and a GCN backbone, but no bidirectional temporal mechanism is described. Please clarify what makes the model bidirectional.
  5. [Sec. 5.3] The text says "the predictions for the cybersecurity industry were also accurate" despite the highest MAE and MAPE among the three sectors; this statement should be tempered or justified with a comparison to a baseline.

Circularity Check

1 steps flagged · score 6.0 of 10

Factor-importance claim is circular: GeoShapley explains the model's predicted changes, not observed industry cluster formation.

  1. fitted input called prediction [Section 5.4 (Spatial Effects and Features Contribution Analysis); Eq. (12); Abstract]
    "The previously predicted rates of change in visitation volumes for each industry are the dependent variable. ... Fig. 13 presents a summary plot of estimated SHAP values for TCI classification... revealing that geolocation exerts the strongest influence on TCI. ... Our findings reveal that geolocation and education levels are the most significant factors influencing industry cluster formation."

    The factor-importance analysis is not run on observed cluster levels or observed visitor-flow changes; it is run on BiTransGCN's predicted change rates, which are themselves the model's output. Eq. (12) defines the GeoShapley output as a decomposition of the model's prediction y-hat, so the resulting ranking of socioeconomic factors is, by construction, an attribution of the fitted forecaster's output. The abstract and conclusion then re-label this model attribution as 'factors influencing industry cluster formation,' converting the model's own predictions into the empirical evidence for the headline claim.

full rationale

The paper's forecasting chain is not itself circular: BiTransGCN is trained on 2022 weekly visitor flows, evaluated on a held-out test split with metrics in Table 4, and the reported 14.16% average growth is a test-set forecast converted to week-on-week change rates. That part is a legitimate model-based prediction. The circularity is concentrated in the factor-importance claim. Section 5.4 explicitly states that 'the previously predicted rates of change in visitation volumes for each industry' serve as the dependent variable for the XGBoost/GeoShapley analysis, and Eq. (12) shows that GeoShapley values sum to the model's prediction. Thus the finding that 'geolocation and education levels are the most significant factors influencing industry cluster formation' is an explanation of the fitted model's output, not an empirical estimate of what drives observed clustering. This is a load-bearing partial circularity because the abstract and conclusion present the model-attribution result as a discovery about the world. I found no load-bearing self-citation chain or uniqueness argument; the GeoShapley citation [51] is an external methodological reference, and the paper's own self-citations are not central to the derivation.

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

The central claims rest on several unverified premises: SafeGraph's panel is representative of TCI visitor flows; the NAICS mapping in Table 1 correctly defines the TCI ecosystem; visitor flows proxy for cluster strength; and the model's predictions are a valid basis for factor-importance conclusions. The main free parameters are the tuned hyperparameters and the arbitrary clustering-level bin boundaries.

free parameters (5)
  • BiTransGCN hyperparameters (dropout, learning rate, hidden size, weight decay, epochs) = dropout=0.05, lr=0.0001, hidden=128, weight_decay=1e-5, epochs=600
    Chosen by 'extensive testing' on the validation portion of the 2022 data (Sec 5.3); a single configuration is reported with no sensitivity analysis.
  • XGBoost hyperparameters (learning rate, max depth, n_estimators, subsample, colsample) = Tuned via Hyperopt, ranges given in Sec 5.4
    Tuned to minimize RMSE on the same data used for the factor-importance analysis, so the reported importances are conditional on this tuning.
  • Number of clusters K in K-means = 6
    Selected as 'optimal number' but no elbow, silhouette, or other criterion is reported (Sec 3.2).
  • Cluster level bin boundaries = 0, 0.17, 0.34, 0.51, 0.68, 0.84, 1.0
    Arbitrary equal-width binning of normalized visitor flow values; the 'cluster levels' are not a statistical clustering solution (Sec 5.2).
  • Weights for six composite social variables = Not specified
    Each category is a 'weighted calculation of three sub-variables' (Sec 5.4); weights are not reported or justified, so the six final variables are not reproducible.
assumptions (5)
  • domain assumption SafeGraph smartphone-location visitor counts are a representative measure of business visitor flows for the TCI industries.
    All cluster maps, Moran's I, and model predictions depend on this panel being representative; the paper provides no validation against census or other ground-truth flow data (Sec 3.1).
  • domain assumption The NAICS-to-TCI categorization in Table 1 correctly identifies the transportation cybersecurity ecosystem.
    The mapping includes broad codes such as 541690 (other scientific and technical consulting) and 561622 (locksmiths) as 'cybersecurity'; misclassification would propagate into every downstream result (Sec 3.1, Table 1).
  • domain assumption Visitor flow volume is a valid proxy for industry cluster strength and workforce mobility.
    The paper equates visitor flows with clustering without testing the proxy against established cluster measures such as location quotients or employment data (Sec 1, Sec 5.2).
  • domain assumption The model's one-week-ahead predictions are a valid dependent variable for explaining real-world socioeconomic influences.
    The factor-importance analysis regresses the model's predicted change rates on socioeconomic variables (Sec 5.4), assuming the model outputs represent actual future dynamics rather than fitted artifacts.
  • ad hoc to paper Standard scaled dot-product attention (Eq. 2) and GCN propagation (Eq. 9) are valid for this 'BiTransGCN' architecture, and the 'bidirectional' component is well-defined.
    The paper describes standard Transformer and GCN components but never defines a bidirectional mechanism; the 'Bi' label is asserted without a corresponding equation or layer (Sec 4.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Geography of Transportation Cybersecurity: Visitor Flows, Industry Clusters, and Spatial Dynamics." pith.science (2026). https://pith.science/paper/VPGRXUOZ

@misc{pith2026250508822,
  author       = {Pith},
  title        = {Pith review of: The Geography of Transportation Cybersecurity: Visitor Flows, Industry Clusters, and Spatial Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VPGRXUOZ}},
  note         = {Machine review of arXiv:2505.08822}
}
read the original abstract

The rapid evolution of the transportation cybersecurity ecosystem, encompassing cybersecurity, automotive, and transportation and logistics sectors, will lead to the formation of distinct spatial clusters and visitor flow patterns across the US. This study examines the spatiotemporal dynamics of visitor flows, analyzing how socioeconomic factors shape industry clustering and workforce distribution within these evolving sectors. To model and predict visitor flow patterns, we develop a BiTransGCN framework, integrating an attention-based Transformer architecture with a Graph Convolutional Network backbone. By integrating AI-enabled forecasting techniques with spatial analysis, this study improves our ability to track, interpret, and anticipate changes in industry clustering and mobility trends, thereby supporting strategic planning for a secure and resilient transportation network. It offers a data-driven foundation for economic planning, workforce development, and targeted investments in the transportation cybersecurity ecosystem.

Figures

Figures reproduced from arXiv: 2505.08822 by the authors.

Figure 1
Figure 1. Geographic coding and CGB density maps [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Research framework chart 4 Methodology 4.1 Long-term Flow Prediction Based on Deep Learning In this study, historical weekly visitor flow data is utilized to project visitor counts for the upcoming period, aligning with the concept of long-term prediction. The proposed BiTransGCN model ( [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. The structure of BiTransGCN model applied for long-term flow prediction [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: a) Scaled dot-product attention and multi-head self-attention and b) the architecture of trans [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: The structure of GCN with multiple graph convolutional layers [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: The OD map of three industries in TCI visitors [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Time series of visitor flows in three industries in TCI [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: The cluster level of visitor flows in TCI at the CBG (a-c), and the bivariate global Moran’s I [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: The cluster level of visitor flows in TCI at the Destination (a-c) and the bivariate global [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: The training loss of BiTransGCN model in TCI visitor flow of prediction [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: The prediction changes of TCI visitor flows on a week-on-week basis [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: The regression coefficient maps (a-f) between 6 categories of variables and visitor flows [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: The average impact evaluation of social variables on the predicted visitor flows in the TCI [PITH_FULL_IMAGE:figures/full_fig_p021_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 55 canonical work pages

  1. [1]

    Cybersecurity regulatory challenges for connected and automated vehicles–state-of-the-art and future directions

    Shah Khalid Khan, Nirajan Shiwakoti, Peter Stasinopoulos, and Matthew Warren. Cybersecurity regulatory challenges for connected and automated vehicles–state-of-the-art and future directions. Transport policy, 143:58–71, 2023

  2. [2]

    Cybersecurity in Transportation Systems: Policies and Technology Directions

    Ostonya Thomas, M Sabbir Salek, Jean-Michel Tine, Mizanur Rahman, Trayce Hockstad, and Mashrur Chowdhury. Cybersecurity in transportation systems: Policies and technology directions. arXiv preprint arXiv:2501.05356 , 2025

  3. [3]

    Bringing business clusters into the mainstream of economic development

    Stuart A Rosenfeld. Bringing business clusters into the mainstream of economic development. European planning studies, 5(1):3–23, 1997

  4. [4]

    High-technology development in regional economic growth: Policy implications of dynamic externalities

    Byung-Rok Choi. High-technology development in regional economic growth: Policy implications of dynamic externalities . Routledge, 2019. 22

  5. [5]

    The impact of regional innovative clusters on competitiveness

    Daniela Doina Fundeanu and Cosmin Sandu Badele. The impact of regional innovative clusters on competitiveness. Procedia-Social and Behavioral Sciences, 124:405–414, 2014

  6. [6]

    Global value chains from an evolutionary economic geography perspective: a research agenda

    Ron Boschma. Global value chains from an evolutionary economic geography perspective: a research agenda. Area Development and Policy , 7(2):123–146, 2022

  7. [7]

    Inter-industry labor flows

    Frank MH Neffke, Anne Otto, and Antje Weyh. Inter-industry labor flows. Journal of Economic Behavior & Organization , 142:275–292, 2017

  8. [8]

    The social consequences of transport decision-making: clarifying concepts, synthesising knowledge and assessing implications

    Peter Jones and Karen Lucas. The social consequences of transport decision-making: clarifying concepts, synthesising knowledge and assessing implications. Journal of transport geography, 21:4– 16, 2012

Show all 61 references
  1. [9]

    Digital entrepreneurial ecosystems: A sys- tematic literature review

    Melissa Bejjani, Lutz G¨ ocke, and Matthias Menter. Digital entrepreneurial ecosystems: A sys- tematic literature review. Technological Forecasting and Social Change, 189:122372, 2023

  2. [10]

    Regional resilience and the network structure of inter-industry labour flows

    Zolt´ an Elekes, Gerg˝ o T´ oth, and Rikard Eriksson. Regional resilience and the network structure of inter-industry labour flows. Regional studies, 58(12):2307–2321, 2024

  3. [11]

    Principles of economics, 8-(edition

    Alfred Marshall. Principles of economics, 8-(edition. Londres, Editions Macmillan , 1920

  4. [12]

    The new argonauts: Regional advantage in a global economy

    AnnaLee Saxenian. The new argonauts: Regional advantage in a global economy . Harvard Uni- versity Press, 2006

  5. [13]

    Sticky places in slippery space: a typology of industrial districts

    Ann Markusen. Sticky places in slippery space: a typology of industrial districts. In Economy, pages 177–197. Routledge, 2017

  6. [14]

    The social structure of entrepreneurial activity: Geographic concentration of footwear production in the united states, 1940–1989

    Olav Sorenson and Pino G Audia. The social structure of entrepreneurial activity: Geographic concentration of footwear production in the united states, 1940–1989. american Journal of Soci- ology, 106(2):424–462, 2000

  7. [15]

    Evolving connection of transit, ag- glomeration, and growth of high-technology business clusters

    Glen Weisbrod, Chandler Duncan, and Susan Jones Moses. Evolving connection of transit, ag- glomeration, and growth of high-technology business clusters. Transportation Research Record, 2452(1):11–18, 2014

  8. [16]

    The comparative advantage of nations, 1990

    Michael Porter. The comparative advantage of nations, 1990

  9. [17]

    The economic geography of talent

    Richard Florida. The economic geography of talent. Annals of the Association of American geographers, 92(4):743–755, 2002

  10. [18]

    Spatial divisions of labour: Social structures and the geography of production

    Doreen Massey. Spatial divisions of labour: Social structures and the geography of production . Bloomsbury Publishing, 1995

  11. [19]

    The regional world: territorial development in a global economy

    Michael Storper. The regional world: territorial development in a global economy . Guilford press, 1997

  12. [20]

    Spatial data and analytical approaches

    Rachel A Rosenfeld and Katherine J Curtis. Spatial data and analytical approaches. In Interna- tional Handbook of Population and Environment , pages 111–137. Springer, 2022

  13. [21]

    The colocation quotient: A new measure of spatial association between categorical subsets of points

    Timothy F Leslie and Barry J Kronenfeld. The colocation quotient: A new measure of spatial association between categorical subsets of points. Geographical Analysis, 43(3):306–326, 2011

  14. [22]

    Detecting industry clusters from the bottom up based on co-location patterns mining: A case study in dongguan, china

    Zihui Liu, Xinyue Chen, Weipan Xu, Yimin Chen, and Xun Li. Detecting industry clusters from the bottom up based on co-location patterns mining: A case study in dongguan, china. Environment and Planning B: Urban Analytics and City Science , 48(9):2827–2841, 2021

  15. [23]

    Spatial spillovers of ag- glomeration economies and productivity in the tourism industry: The case of the uk

    Yoo Ri Kim, Allan M Williams, Sangwon Park, and Jason Li Chen. Spatial spillovers of ag- glomeration economies and productivity in the tourism industry: The case of the uk. Tourism management, 82:104201, 2021

  16. [24]

    Big data analytics, artificial in- telligence, machine learning, internet of things, and blockchain for enhanced business intelligence

    Mallikarjuna Paramesha, Nitin Liladhar Rane, and Jayesh Rane. Big data analytics, artificial in- telligence, machine learning, internet of things, and blockchain for enhanced business intelligence. Partners Universal Multidisciplinary Research Journal , 1(2):110–133, 2024. 23

  17. [25]

    Vector autoregressive models for multivariate time series

    Eric Zivot and Jiahui Wang. Vector autoregressive models for multivariate time series. Modeling financial time series with S-PLUS ®, pages 385–429, 2006

  18. [26]

    Time series forecasting using a hybrid arima and neural network model

    G Peter Zhang. Time series forecasting using a hybrid arima and neural network model. Neuro- computing, 50:159–175, 2003

  19. [27]

    Short-term traffic flow forecasting with spatial-temporal correla- tion in a hybrid deep learning framework

    Yuankai Wu and Huachun Tan. Short-term traffic flow forecasting with spatial-temporal correla- tion in a hybrid deep learning framework. arXiv preprint arXiv:1612.01022 , 2016

  20. [28]

    Learning dynamic and hierarchical traffic spa- tiotemporal features with transformer

    Haoyang Yan, Xiaolei Ma, and Ziyuan Pu. Learning dynamic and hierarchical traffic spa- tiotemporal features with transformer. IEEE Transactions on Intelligent Transportation Systems, 23(11):22386–22399, 2021

  21. [29]

    A multi- head attention-based transformer model for traffic flow forecasting with a comparative analysis to recurrent neural networks

    Selim Reza, Marta Campos Ferreira, Jos´ e JM Machado, and Jo˜ ao Manuel RS Tavares. A multi- head attention-based transformer model for traffic flow forecasting with a comparative analysis to recurrent neural networks. Expert Systems with Applications , 202:117275, 2022

  22. [30]

    Dnn-based prediction model for spatio-temporal data

    Junbo Zhang, Yu Zheng, Dekang Qi, Ruiyuan Li, and Xiuwen Yi. Dnn-based prediction model for spatio-temporal data. In Proceedings of the 24th ACM SIGSPATIAL international conference on advances in geographic information systems , pages 1–4, 2016

  23. [31]

    Augmentation-free graph contrastive learning of invariant-discriminative representations

    Haifeng Li, Jun Cao, Jiawei Zhu, Qinyao Luo, Silu He, and Xuying Wang. Augmentation-free graph contrastive learning of invariant-discriminative representations. IEEE Transactions on Neural Networks and Learning Systems , 2023

  24. [32]

    Identification and factor analysis of rocky desertification severity levels in large-scale karst areas based on deep learning image segmentation

    Yuhao Wang, Xianghong Tang, Yong Huang, Jing Yang, and Jianguang Lu. Identification and factor analysis of rocky desertification severity levels in large-scale karst areas based on deep learning image segmentation. Ecological Indicators, 167:112565, 2024

  25. [33]

    Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting

    Bing Yu, Haoteng Yin, and Zhanxing Zhu. Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting. arXiv preprint arXiv:1709.04875 , 2017

  26. [34]

    T-gcn: A temporal graph convolutional network for traffic prediction

    Ling Zhao, Yujiao Song, Chao Zhang, Yu Liu, Pu Wang, Tao Lin, Min Deng, and Haifeng Li. T-gcn: A temporal graph convolutional network for traffic prediction. IEEE transactions on intelligent transportation systems , 21(9):3848–3858, 2019

  27. [35]

    Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting

    Chao Song, Youfang Lin, Shengnan Guo, and Huaiyu Wan. Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting. In Proceedings of the AAAI conference on artificial intelligence , volume 34, pages 914–921, 2020

  28. [36]

    Spatio-temporal graph structure learning for traffic forecasting

    Qi Zhang, Jianlong Chang, Gaofeng Meng, Shiming Xiang, and Chunhong Pan. Spatio-temporal graph structure learning for traffic forecasting. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 1177–1185, 2020

  29. [37]

    Stgc- gnns: A gnn-based traffic prediction framework with a spatial–temporal granger causality graph

    Silu He, Qinyao Luo, Ronghua Du, Ling Zhao, Guangjun He, Han Fu, and Haifeng Li. Stgc- gnns: A gnn-based traffic prediction framework with a spatial–temporal granger causality graph. Physica A: Statistical Mechanics and its Applications , 623:128913, 2023

  30. [38]

    St-dagcn: A spatiotemporal dual adaptive graph convolutional network model for traffic prediction

    Yutian Liu, Tao Feng, Soora Rasouli, and Melvin Wong. St-dagcn: A spatiotemporal dual adaptive graph convolutional network model for traffic prediction. Neurocomputing, 601:128175, 2024

  31. [39]

    Sat-gcn: Self-attention graph convolutional network-based 3d object detection for au- tonomous driving

    Li Wang, Ziying Song, Xinyu Zhang, Chenfei Wang, Guoxin Zhang, Lei Zhu, Jun Li, and Huap- ing Liu. Sat-gcn: Self-attention graph convolutional network-based 3d object detection for au- tonomous driving. Knowledge-Based Systems, 259:110080, 2023

  32. [40]

    Attention is all you need

    Vaswani Ashish. Attention is all you need. Advances in neural information processing systems , 30:I, 2017

  33. [41]

    Semi-supervised classification with graph convolutional net- works

    Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional net- works. arXiv preprint arXiv:1609.02907 , 2016

  34. [42]

    Data clustering: application and trends

    Gbeminiyi John Oyewole and George Alex Thopil. Data clustering: application and trends. Artificial intelligence review, 56(7):6439–6475, 2023. 24

  35. [43]

    Why do industries cluster

    Peter B Doeringer and David G Terkla. Why do industries cluster. Business networks: Prospects for regional development, pages 175–189, 1996

  36. [44]

    How bad is crime for business? evidence from consumer behavior

    Hao Fe and Viviane Sanfelice. How bad is crime for business? evidence from consumer behavior. Journal of urban economics , 129:103448, 2022

  37. [45]

    Industry clusters and employment outcomes in washington state

    Colleen K Chrisinger, Christopher S Fowler, and Rachel Garshick Kleit. Industry clusters and employment outcomes in washington state. Economic Development Quarterly , 29(3):199–210, 2015

  38. [46]

    The effects of industry structure on economic exposure

    Richard C Marston. The effects of industry structure on economic exposure. Journal of Interna- tional Money and Finance , 20(2):149–164, 2001

  39. [47]

    The effect of regional economic clusters on housing price

    Jiaying Kou, Jiahua Du, Xiaoming Fu, Geordie Z Zhang, Hua Wang, and Yanchun Zhang. The effect of regional economic clusters on housing price. In Databases Theory and Applications: 32nd Australasian Database Conference, ADC 2021, Dunedin, New Zealand, January 29–February 5, 202...

  40. [48]

    A cluster separation measure

    Davies Dl. A cluster separation measure. IEEE Trans Pattern Anal Mach Intell , 1:224–227, 1979

  41. [49]

    The analysis of spatial association by use of distance statistics

    Arthur Getis and J Keith Ord. The analysis of spatial association by use of distance statistics. Geographical analysis, 24(3):189–206, 1992

  42. [50]

    Joint shapley values: a measure of joint feature importance

    Chris Harris, Richard Pymar, and Colin Rowat. Joint shapley values: a measure of joint feature importance. arXiv preprint arXiv:2107.11357 , 2021

  43. [51]

    Geoshapley: A game theory approach to measuring spatial effects in machine learning models

    Ziqi Li. Geoshapley: A game theory approach to measuring spatial effects in machine learning models. Annals of the American Association of Geographers , 114(7):1365–1385, 2024

  44. [52]

    Determinants of supplier plant location: Evidence from the auto industry

    Thomas Klier. Determinants of supplier plant location: Evidence from the auto industry. Eco- nomic Perspectives, 3:2–15, 2005

  45. [53]

    Agglomeration and industrial location: An econometric analysis of japanese- affiliated manufacturing establishments in automotive-related industries

    Richard Florida et al. Agglomeration and industrial location: An econometric analysis of japanese- affiliated manufacturing establishments in automotive-related industries. Journal of urban eco- nomics, 36(1):23–41, 1994

  46. [54]

    Intra-industry trade: Transportation needs of growing texas industries

    Robert Harrison. Intra-industry trade: Transportation needs of growing texas industries. 2015

  47. [55]

    Recent trends in manufacturing innovation policy for the automotive sector: A survey of the united states, mexico, european union, germany and spain

    Patrick Galvin, Elena Goracinova, and David A Wolfe. Recent trends in manufacturing innovation policy for the automotive sector: A survey of the united states, mexico, european union, germany and spain. The global automotive industry , pages 53–66, 2015

  48. [56]

    Xgboost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages 785–794, 2016

  49. [57]

    Knowledge flows through social networks in a cluster: Comparing uni- versity and industry links

    Christian R Østergaard. Knowledge flows through social networks in a cluster: Comparing uni- versity and industry links. Structural Change and Economic Dynamics , 20(3):196–210, 2009

  50. [58]

    Logistics cluster and its future development: A comprehensive research review

    Sijing Liu, Nannan He, Xindan Cao, Guoqi Li, and Ming Jian. Logistics cluster and its future development: A comprehensive research review. Transportation Research Part E: Logistics and Transportation Review, 168:102974, 2022

  51. [59]

    The co-location of innovation and production in clusters

    Mercedes Delgado. The co-location of innovation and production in clusters. Industry and Inno- vation, 27(8):842–870, 2020

  52. [60]

    Logistics agglomeration and logistics productivity in the usa

    Bowen Sun, Haomin Li, and Qiuyun Zhao. Logistics agglomeration and logistics productivity in the usa. The Annals of Regional Science , 61:273–293, 2018

  53. [61]

    Impacts of logistics sprawl on the urban environment and logistics: Taxonomy and review of literature

    Khalid Aljohani and Russell G Thompson. Impacts of logistics sprawl on the urban environment and logistics: Taxonomy and review of literature. Journal of Transport Geography, 57:255–263, 2016. 25

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.