Pith. sign in

REVIEW 3 major objections 6 minor 32 references

Hierarchical Spatio-Temporal Transformer for Coherent Emergency Department Forecasting

T0 review · 3 major / 6 minor · reviewed 2026-07-30 · grok-4.5

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

Pith's one-line read A single top-down Transformer can forecast emergency-department demand at hospital, region, and nation at once, keeping the numbers consistent and cutting error by about a third.

desk verdict Solid applied hierarchical ED forecasting paper with a real national dataset and honest accuracy–coherence results; main gap is missing reconciled deep baselines, not the core claim. read the letter →

arxiv 2607.27106 v1 pith:DVZQB72C submitted 2026-07-29 cs.LG

Filipa Lino , Bárbara Tavares , Carlos Santiago , Cláudia Soares , Manuel Marques This is my paper · ORCID
classification cs.LG
keywords EmergencyDepartmentForecastingHierarchicalTimeSeriesTransformersSpatio-TemporalModelingForecastCoherenceTemporalFusionTransformerHealthcaredemand
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

Emergency departments need demand forecasts at three linked scales—individual hospitals for staffing, regions for coordination, and the nation for capacity—yet most models predict each scale separately and the numbers do not add up. This paper introduces HierSTT, one end-to-end model that forecasts all three levels together. It first predicts national demand with a Temporal Fusion Transformer, then conditions regional and hospital spatio-temporal Transformers on those higher-level forecasts, and trains with a loss that penalizes inconsistent aggregates. On a new Portuguese dataset of 81 hospitals in five regions, the model cuts average weighted absolute percentage error by roughly 32% versus the best non-hierarchical deep baseline, beats classical post-hoc reconciliation methods, and keeps cross-level predictions nearly coherent. A sympathetic reader cares because incoherent multi-level forecasts undermine real staffing and policy decisions; a single model that is both more accurate and internally consistent is directly usable for planning.

What carries the argument

HierSTT: top-down conditioning (national forecast guides regional decoding; regional forecast guides hospital decoding) together with a coherence-aware loss that penalizes mismatches between direct higher-level predictions and sums of lower-level predictions, without hard constraints.

What would settle it

Train the same deep baselines (e.g., N-BEATS, TFT, Seq2Seq) independently, then apply standard bottom-up/top-down/middle-out or MinT reconciliation, and test whether average WAPE and ground-truth hierarchical aggregation error still trail HierSTT by a large margin on the held-out Portuguese test window.

Watch

Extended reading notes

Core claim

HierSTT shows that jointly training a top-down hierarchical Transformer—national TFT plus regional and hospital spatio-temporal encoder-decoders conditioned on higher-level forecasts, plus a soft coherence loss—produces more accurate ED visit forecasts at every level and near-coherent aggregates than independent deep models or classical hierarchical reconciliation on a nationwide 81-hospital hierarchy.

Load-bearing premise

The comparison treats independent deep models and reconciliation applied only to classical statistical forecasts as a fair test; it does not check whether reconciling strong deep base forecasts would close most of the gap.

Editorial extensions

If this is right

  • Hospitals, regions, and national planners can share one model whose hospital forecasts sum to the regional and national numbers they already use.
  • Heterogeneous covariates (local operations vs regional weather/AQI vs national mortality) can be fused without forcing a common feature set.
  • Soft coherence training can replace or reduce reliance on post-hoc reconciliation for multi-level healthcare demand.
  • The released 81-hospital Portuguese ED dataset becomes a benchmark for hierarchical spatio-temporal forecasting with real closures and level-specific missingness.

Reading between the lines

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

  • The same top-down conditioning plus coherence loss pattern could transfer to other administrative hierarchies (beds, ambulances, primary-care load) where lower units must add to upper budgets.
  • If reconciled deep bases close little of the gap, the gain is mainly from learning cross-level attention during training rather than from architecture size alone.
  • Near-zero prediction HAgE with non-zero ground-truth HAgE suggests the model is learning a coherent internal world that still has residual bias versus reality—useful for stress-testing planning scenarios.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes HierSTT, an end-to-end hierarchical Transformer for multi-level Emergency Department (ED) demand forecasting over a 28-day horizon from 42-day histories. A Temporal Fusion Transformer produces national forecasts; spatio-temporal Transformer encoder–decoders then produce regional and hospital forecasts conditioned top-down on higher-level predictions (Eqs. 4–6). A soft coherence loss (Eqs. 7–8) penalizes inconsistencies among prediction aggregates. The authors introduce a Portuguese nationwide dataset (81 hospitals, 5 RHAs, heterogeneous covariates) and evaluate accuracy (MAE/RMSE/WAPE) and hierarchical coherence (Pred./G.T. HAgE). They report ~32% lower average WAPE than the best non-hierarchical deep baseline, better accuracy–coherence trade-offs than classical Nixtla reconciliation of Naïve/ARIMA/ETS, and near-coherent cross-level forecasts, with α ablation and multi-seed reporting.

Significance. If the empirical claims hold under stronger hierarchical deep baselines, the work is a solid applied contribution: joint multi-level ED forecasting with learned coherence, a real multi-admin Portuguese dataset with level-heterogeneous covariates, and an evaluation protocol that reports both Pred. and G.T. HAgE rather than accuracy alone. Strengths include chronological splits, log+RobustScaler pipeline with inverse-domain coherence, six-seed mean±std, α ablation (Table 2), full per-level tables, and a public code link. The architecture preserves standalone national TFT performance while improving lower levels, which is a useful internal control. The main significance limit is baseline scope rather than internal inconsistency: hierarchical advantage is shown against independent deep models and classical reconciled stats, not against reconciled or lightly constrained strong deep bases.

major comments (3)
  1. [§5.2 Baselines; Tables 3–4] §5.2 and Tables 3–4: hierarchical baselines are only Bottom-Up/Top-Down/Middle-Out reconciliation of Naïve/ARIMA/ETS via Nixtla. No post-hoc reconciliation (or MinT-style) is applied to strong deep bases (N-BEATS, TFT, Seq2seq, FA-Transformer), nor are simple hierarchical deep variants (shared backbone + hard aggregation layer, bottom-up deep, or Mancuso-style reconciliation layers) reported. The abstract’s claim of outperforming “all classical hierarchical reconciliation methods” and the ~32% average-WAPE gain vs “the best non-hierarchical deep learning baseline” therefore leave open how much gain is end-to-end top-down conditioning + L_coh versus any hierarchical treatment of a strong base. At minimum, add Bottom-Up/MinT (or equivalent) on N-BEATS and hospital-level TFT/Seq2seq forecasts, and discuss residual gap.
  2. [§2.3; §5.2] Related Work §2.3 cites Mancuso et al. and NeuralReconciler as integrating reconciliation into training, but experiments never include these (or simplified analogues) as deep hierarchical baselines. Given the paper’s claim that post-hoc methods “fail to exploit cross-level dependencies during learning,” a direct comparison—or an explicit limitation paragraph stating they were out of scope and why—is load-bearing for the novelty/advantage narrative. Without it, the hierarchical DL contribution is only partially stress-tested.
  3. [Abstract; §6.1; Table 3] Abstract and §6.1 state ~32% average WAPE reduction vs the best non-hierarchical DL baseline and strong regional gains (e.g., ~38% MAE vs Naïve). Table 3 shows HierSTT hospital WAPE 10.8% vs N-BEATS 14.1% (~23% relative) and national WAPE tied with TFT at 5.4%; the “average WAPE” aggregation across levels is not defined in the main text (equal weight per level? demand-weighted?). Please define the average explicitly (formula + which models enter the 32% comparison) so the headline number is reproducible from Table 3.
minor comments (6)
  1. [Fig. 4; §5.2] Fig. 4 legend lists “FA-Transformer” while §5.2 baselines name “Transformer” and the extended table explains FA-Transformer as future-aware; unify naming in main-text baseline list and figure captions.
  2. [§4.1 Eq. (1)] Eq. (1) writes the national identity twice (sum over hospitals and sum over regions); harmless but redundant—tighten.
  3. [§6.1; Table 3; Fig. 11] Hospital-level RMSE for HierSTT is slightly worse than N-BEATS (Table 3) while MAE/WAPE improve; a one-sentence discussion of peak underestimation (also visible in Fig. 11) would help readers interpret the accuracy–coherence trade-off.
  4. [§5 Dataset and Splits] Data period excludes pre-Aug 2021 for COVID; briefly note whether any remaining structural breaks (e.g., policy changes) were checked in the 2021–2024 window.
  5. [Throughout] Typos/style: “W APE” spacing in places; “F orecasting” line-break artifacts in headings; “have sought” agreement in §2.3; arXiv ID/date formatting in the header is fine for preprint but clean for camera-ready.
  6. [Appendix B; §5.1] Supplementary α analysis (Table 2) is valuable; consider moving the effective-ratio argument (Eq. 10 in appendix) to a short main-text remark so readers understand why α=0.3 is not an arbitrary soft weight.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: standard supervised multi-level forecasting with soft coherence penalty and held-out metrics.

full rationale

HierSTT’s claims are empirical, not algebraic identities. National/regional/hospital forecasts are produced by TFT and spatio-temporal encoder–decoders and scored with MAE/RMSE/WAPE against held-out ground truth after inverse transforms (Eq. 9, §5–6, Tables 3–4). The coherence term L_coh (Eqs. 7–8) only compares cross-level predictions to each other with a soft weight α; it does not define accuracy, and the α ablation (Table 2) shows Pred. HAgE falls and accuracy degrades as α rises—so coherence is learned, not forced to zero by construction. Top-down conditioning (Eqs. 4–6) is an architectural choice, not a fit renamed as prediction. The sole author-overlapping citation (Caldas & Soares TFT on Portuguese RHAs) is background related work, not a uniqueness theorem or load-bearing premise. No step reduces a reported WAPE/HAgE gain to a fitted input or self-definition. Baseline-design limits are scope issues, not circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 3 invented entities

Load-bearing content is empirical ML: hierarchy-as-sums, top-down conditioning, soft coherence weighting, and standard deep sequence modules. Free parameters are training/design choices (α, window lengths, widths). No new physical entities. Domain assumptions are the Portuguese administrative hierarchy and post-COVID stationarity of the modeling window.

free parameters (4)
  • coherence weight α = 0.3
    Balances direct Smooth-L1 losses vs L_coh; selected as 0.3 from ablation for best accuracy–coherence trade-off (Appendix B).
  • encoder/decoder windows Tin, Tout = Tin=42 days, Tout=28 days
    Fixed lookback and horizon used for all models; chosen for weekly seasonality and operational planning, not derived.
  • model width and depth (d_model, L, heads, dropout, LR) = d_model=128, L=2, nhead=4, peak lr=3e-4
    Architecture and optimizer hyperparameters set by authors (d_model=128, L=2, 4 heads, dropout 0.1–0.2, AdamW 3e-4 peak, etc.).
  • train/val/test chronological cutoffs = 713 / 140 / 140 days
    Data after excluding early COVID; cutoffs 15 Jul 2023 and 2 Dec 2023 define the evaluation that supports the 32% claim.
assumptions (5)
  • domain assumption Regional and national series equal exact sums of member hospital (and region) series at every t (Eq. 1).
    Defines hierarchical coherence targets; standard HTS assumption, invoked throughout problem formulation and L_coh.
  • ad hoc to paper Top-down factorization: national forecast first, then regions conditioned on national, hospitals conditioned on parent region (Eqs. 4–6).
    Architectural choice, not forced by the hierarchy; alternatives (bottom-up, joint bidirectional) are not jointly trained here.
  • ad hoc to paper Soft coherence via Smooth-L1 on prediction aggregates is sufficient; hard projection layers are unnecessary.
    §4.4 loss design; Pred. HAgE is reduced but not forced to zero.
  • domain assumption Post-August-2021 series are suitable for supervised multi-horizon learning after log1p and RobustScaler.
    §5 excludes COVID-shifted demand; preprocessing stabilizes cross-level scales before inverse-transform coherence checks.
  • standard math Standard sequence-model building blocks (TFT, multi-head attention, LSTM VSNs) are valid function classes for this data.
    Imported from cited forecasting literature without re-proof.
invented entities (3)
  • HierSTT architecture (national TFT + regional/hospital ST Transformer encoder–decoders with higher-level forecast injection into decoder queries)
    purpose: Joint multi-level ED forecasting with learned cross-level conditioning and coherence.
    Named system composing known modules; evaluated only inside this paper’s experiments.
  • Portuguese multi-level ED dataset (81 hospitals, 5 RHAs, heterogeneous covariates) independent evidence
    purpose: Enable and benchmark hierarchical ED forecasting under real covariate asymmetry.
    New resource assembled from public SNS and environmental sources; independent of the model once released.
  • Hierarchical Aggregation Error (HAgE) reporting protocol (Pred. vs G.T.) independent evidence
    purpose: Measure prediction-side incoherence and error propagation separately from point accuracy.
    Metric framing built from WAPE on aggregates; useful but not a physical entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hierarchical Spatio-Temporal Transformer for Coherent Emergency Department Forecasting." pith.science (2026). https://pith.science/paper/DVZQB72C

@misc{pith2026260727106,
  author       = {Pith},
  title        = {Pith review of: Hierarchical Spatio-Temporal Transformer for Coherent Emergency Department Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DVZQB72C}},
  note         = {Machine review of arXiv:2607.27106}
}
read the original abstract

Emergency Departments (EDs) are critical access points in healthcare systems, yet they face persistent pressure from unpredictable patient demand, seasonal surges, and non-urgent visits. Effective ED planning requires forecasts at multiple decision-making levels: hospitals need local demand estimates for staffing and bed management, regions require forecasts to coordinate healthcare units, and national authorities need system-wide projections for capacity planning. However, most existing approaches forecast ED demand independently at a single level, ignoring the hierarchy linking hospitals, regions, and national systems. This can produce incoherent predictions, where hospital-level forecasts do not aggregate consistently to regional or national demand. We propose HierSTT, a hierarchical Transformer-based framework for coherent multi-level ED forecasting. HierSTT jointly predicts hospital, regional, and national level demand in a single end-to-end model. A Temporal Fusion Transformer captures national dynamics, while spatio-temporal Transformer encoder-decoder modules model regional and hospital demand conditioned on higher-level forecasts. A coherence-aware loss penalizes cross-level inconsistencies during training. We further introduce a nationwide Portuguese ED dataset covering 81 hospitals across 5 regional health administrations, with heterogeneous covariates at each level. Experiments show that HierSTT reduces average WAPE by 32\% relative to the best non-hierarchical deep learning baseline and outperforms all classical hierarchical reconciliation methods, while producing near-coherent predictions across levels. Additional resources associated with this work are available at https://github.com/FilipaLino/HierSTT.

Figures

Figures reproduced from arXiv: 2607.27106 by the authors.

Figure 1
Figure 1. Geographical and hierarchical organization of the ED forecasting dataset. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the HierSTT. A TFT models national-level ED demand, [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Spatio-temporal Transformer encoder–decoder used for regional and hos [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Mean MAE vs. mean RMSE for the hospital, regional, and national levels. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Trade-off between forecasting accuracy (mean WAPE) and hierarchi [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Daily hospital-level M1 time series within each region acrosss the full [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]
Figure 7
Figure 7. Figure 7: Hospital-level M1 series over the modeling period for Region 0 (a) and [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: Sliding window strategy used to generate samples. Each sample consists [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Example of the national-level forecast for the test sample with the low [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]
Figure 10
Figure 10. Figure 10: Regional-level forecasts for the test sample with the lowest WAPE in [PITH_FULL_IMAGE:figures/full_fig_p026_10.png]
Figure 11
Figure 11. Figure 11: Hospital-level forecasts for representative hospitals selected from each [PITH_FULL_IMAGE:figures/full_fig_p027_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 1 linked inside Pith

  1. [1]

    Agency, U.S.E.P.: Technical assistance document for the reporting of daily air quality the air quality index (aqi)

  2. [2]

    Expert Systems with Applications 240, 122496 (2024)

    Álvarez-Chaves, H., Maseda-Zurdo, I., Muñoz, P., R-Moreno, M.D.: Evaluating the impact of exogenous variables for patients forecasting in an emergency department using attention neural networks. Expert Systems with Applications 240, 122496 (2024)

    2024

  3. [3]

    Academic Emergency Medicine 8(1), 48–53 (2001)

    Batal, H., Tench, J., McMillan, S., Adams, J., Mehler, P.S.: Predicting patient visits to an urgent care clinic using calendar variables. Academic Emergency Medicine 8(1), 48–53 (2001)

    2001

  4. [4]

    John Wiley & Sons (2015)

    Box, G.E., Jenkins, G.M., Reinsel, G.C., Ljung, G.M.: Time series analysis: fore- casting and control. John Wiley & Sons (2015)

    2015

  5. [5]

    Emergency Medicine Jour- nal 29(5), 358–365 (2012)

    Boyle, J., Jessup, M., Crilly, J., Green, D., Lind, J., Wallis, M., Miller, P., Fitzger- ald, G.: Predicting emergency department admissions. Emergency Medicine Jour- nal 29(5), 358–365 (2012)

    2012

  6. [6]

    In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases

    Caldas, F.M., Soares, C.: A temporal fusion transformer for long-term explainable prediction of emergency department overcrowding. In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases. pp. 71–88. Springer (2022)

    2022

  7. [7]

    Journal of Business & Economic Statistics 16(2), 127–152 (1998)

    Findley, D.F., Monsell, B.C., Bell, W.R., Otto, M.C., Chen, B.C.: New capabilities and methods of the x-12-arima seasonal-adjustment program. Journal of Business & Economic Statistics 16(2), 127–152 (1998)

    1998

  8. [8]

    OTexts (2018)

    Hyndman, R.J., Athanasopoulos, G.: Forecasting: principles and practice. OTexts (2018)

    2018

Show all 32 references
  1. [9]

    International Journal of forecasting 18(3), 439–454 (2002)

    Hyndman, R.J., Koehler, A.B., Snyder, R.D., Grose, S.: A state space frame- work for automatic forecasting using exponential smoothing methods. International Journal of forecasting 18(3), 439–454 (2002)

    2002

  2. [10]

    IEEE Transactions on Big Data 11(2), 372–387 (2024)

    Ji, J., He, J., Lei, M., Wang, M., Tang, W.: Spatio-temporal transformer network for weather forecasting. IEEE Transactions on Big Data 11(2), 372–387 (2024)

    2024

  3. [11]

    International Journal of Data Science 5(1), 1–25 (2020)

    Kadri, F., Abdennbi, K.: Rnn-based deep-learning approach to forecasting hospital system demands: application to an emergency department. International Journal of Data Science 5(1), 1–25 (2020)

    2020

  4. [12]

    Expert Systems with Applications 37(2), 1784–1789 (2010)

    Kayacan, E., Ulutas, B., Kaynak, O.: Grey system theory-based models in time series prediction. Expert Systems with Applications 37(2), 1784–1789 (2010)

    2010

  5. [13]

    Journal of the Operational Research Society 74, 1–20 (09 2022)

    Makridakis, S., Spiliotis, E., Assimakopoulos, V., Semenoglou, A.A., Mulder, G., Nikolopoulos, K.: Statistical, machine learning and deep learning forecasting meth- ods: Comparisons and ways forward. Journal of the Operational Research Society 74, 1–20 (09 2022)

    2022

  6. [14]

    Expert Systems with Applications 182, 115102 (2021) 16 F

    Mancuso, P., Piccialli, V., Sudoso, A.M.: A machine learning approach for fore- casting hierarchical time series. Expert Systems with Applications 182, 115102 (2021) 16 F. Lino et al

    2021

  7. [15]

    Energies 15(7), 2623 (2022)

    Masood, Z., Gantassi, R., Choi, Y.: A multi-step time-series clustering-based seq2seq lstm learning for a single household electricity load forecasting. Energies 15(7), 2623 (2022)

    2022

  8. [16]

    Statistics in Medicine 7(10), 1061–1072 (1988)

    Milner, P.C.: Forecasting the demand on accident and emergency departments in health districts in the trent region. Statistics in Medicine 7(10), 1061–1072 (1988)

    1988

  9. [17]

    Computer methods and programs in biomedicine 267, 108808 (2025)

    Ming, C., Lee, G.J., Teo, Y.N., Teo, Y.H., Zhou, X., Ho, E.S., Toh, E.M., Ong, M.E.H., Tan, B.Y., Ho, A.F.: Deep learning modelling to forecast emergency de- partment visits using calendar, meteorological, internet search data and stock mar- ket price. Computer methods and pro...

    2025

  10. [18]

    Artificial Intelligence in Emergency Medicine 1, 100006 (2026)

    Ming, C., Leung, K.B., Shen, Y., Ho, A.F.: Transformers outperform traditional forecasting models and perform comparably to recurrent neural networks in the prediction of emergency department visits using calendar and meteorological data. Artificial Intelligence in Emergency Me...

    2026

  11. [19]

    Jour- nal of Archives in Military Medicine 12(1), e140888 (2024)

    Moosavi Kashani, S., Zargar Balaye Jame, S., Markazi-Moghaddam, N., Om- rani Nava, A.: Comparison of long short-term memory and convolutional neural network models for emergency department patients arrival daily forecasting. Jour- nal of Archives in Military Medicine 12(1), e1...

    2024

  12. [20]

    Journal of Emergencies Trauma and Shock 12, 268–273 (11 2019)

    Moreno, A., Muñoz, O.: Application of queuing theory to optimize the triage pro- cess in a tertiary emergency care (er) department. Journal of Emergencies Trauma and Shock 12, 268–273 (11 2019)

    2019

  13. [21]

    International Journal of Forecasting 38(4), 14481459 (Oct 2022)

    Nasios, I., Vogklis, K.: Blending gradient boosted trees and neural networks for point and probabilistic forecasting of hierarchical time series. International Journal of Forecasting 38(4), 14481459 (Oct 2022)

    2022

  14. [22]

    Olivares, K.G., Garza, F., Luo, D., Challú, C., Mergenthaler, M., Taieb, S.B., Wickramasuriya, S.L., Dubrawski, A.: Hierarchicalforecast: A reference framework for hierarchical forecasting in python (2023)

    2023

  15. [23]

    arxiv 2019

    Oreshkin, B.N., Carpov, D., Chapados, N., Bengio, Y.: N-beats: Neural basis ex- pansion analysis for interpretable time series forecasting. arxiv 2019. arXiv preprint arXiv:1905.10437 (2019)

    2019 arXiv

  16. [24]

    Pulkkinen, E.: Forecasting emergency department arrivals with neural networks. B.S. thesis (2020)

    2020

  17. [25]

    Journal of personalized medicine 12(2), 279 (2022)

    Savioli, G., Ceresa, I.F., Gri, N., Bavestrello Piccini, G., Longhitano, Y., Zanza, C., Piccioni, A., Esposito, C., Ricevuti, G., Bressan, M.A.: Emergency department overcrowding: understanding the factors to find corresponding solutions. Journal of personalized medicine 12(2),...

    2022

  18. [26]

    The International Journal of Health Planning and Management 38(4), 904–917 (2023)

    Silva, E., Pereira, M.F., Vieira, J.T., Ferreira-Coimbra, J., Henriques, M., Ro- drigues, N.F.: Predicting hospital emergency department visits accurately: A sys- tematic review. The International Journal of Health Planning and Management 38(4), 904–917 (2023)

    2023

  19. [27]

    Health Economics 29(12), 1744–1763 (2020)

    Turner, A.J., Anselmi, L., Lau, Y.S., Sutton, M.: The effects of unexpected changes in demand on the performance of emergency departments. Health Economics 29(12), 1744–1763 (2020)

    2020

  20. [28]

    Advances in neural information pro- cessing systems 30 (2017)

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, Ł., Polosukhin, I.: Attention is all you need. Advances in neural information pro- cessing systems 30 (2017)

    2017

  21. [29]

    In: Proceedings of the 17th ACM International Conference on Web Search and Data Mining

    Wang, S.: Neuralreconciler for hierarchical time series forecasting. In: Proceedings of the 17th ACM International Conference on Web Search and Data Mining. p. 731739. WSDM ’24, Association for Computing Machinery, New York, NY, USA (2024) Hierarchical Spatio-Temporal Transfor...

    2024

  22. [30]

    Water 13(8), 1031 (2021)

    Xu, J., Wang, K., Lin, C., Xiao, L., Huang, X., Zhang, Y.: Fm-gru: A time series prediction method for water quality based on seq2seq framework. Water 13(8), 1031 (2021)

    2021

  23. [31]

    Statistical Papers of George Udny Yule pp

    Yule, G.U.: On a method of investigating periodicities in disturbed series with special reference to wolfers sunspot numbers. Statistical Papers of George Udny Yule pp. 389–420 (1971)

    1971

  24. [32]

    In: 2018 IEEE 3rd Advanced Information Tech- nology, Electronic and Automation Control Conference (IAEAC)

    Zhang, Q., Wang, K., Guo, H., Yang, S., Li, C.: Base on ets model for forcast- ing emergency department visits. In: 2018 IEEE 3rd Advanced Information Tech- nology, Electronic and Automation Control Conference (IAEAC). pp. 2148–2151. IEEE (2018) 18 F. Lino et al. A Analysis of...

    2018

Pith tools

Reviewed July 30, 2026 · model on record in the stance chip above.