Forecasting COVID-19 new cases using NBEATS deep learning and mobility data

Clinical Snapshot

Key Findings

• P Value: Not reported — no formal statistical hypothesis testing for forecast comparison differences is described

• Effect Size: Specific numerical RMSE and MAPE values are referenced in the abstract as consistently lower for N-BEATS, but precise effect sizes and percentage improvements are not reported in the abstract; full quantitative results require access to the complete manuscript tables.

• Primary Outcome: N-BEATS consistently achieved lower RMSE and MAPE than LSTM-Markov across all four countries (US, UK, Russia, Brazil) and all three datasets (Google, Apple, OWID). N-BEATS with mobility covariates outperformed N-BEATS without covariates, indicating incremental value of mobility data integration.

• Nnt Or Sensitivity: Not applicable for a predictive modelling study; no diagnostic accuracy metrics, hazard ratios, or number-needed-to-treat figures are relevant. Forecasting accuracy is expressed solely as RMSE and MAPE without calibration metrics or prediction interval coverage.

• Confidence Interval: Not reported — a critical methodological omission for a forecasting study intended to inform public health decision-making

Clinical Application

Implementation feasibility is uncertain. The study does not report computational requirements, inference latency, or the technical infrastructure needed for real-time deployment. Continuous retraining requirements as pandemic dynamics evolve are not addressed. Integration with existing public health surveillance systems (e.g., national notifiable disease reporting) would require substantial technical and governance investment. Limited direct applicability to Australian practice. Australia's COVID-19 trajectory was shaped by unique factors — island geography, state-based public health responses, National Cabinet coordination, and a phased vaccination rollout under the TGA-approved schedule — that differ substantially from the four countries studied. The Australian Centre for Disease Control (ACDC), established in 2023, and state health departments use established epidemiological modelling frameworks (including Doherty Institute models) that incorporate vaccination coverage, variant-specific parameters, and healthcare capacity data not present in this study. The RACGP and AHPPC would require prospective validation in Australian data before any such tool could inform clinical or policy guidance. PBS and TGA considerations are not relevant to this computational modelling study. Potentially applicable to national-level public health surveillance and pandemic preparedness planning in high-income countries with robust digital mobility data infrastructure. Not validated for sub-national, regional, or local health service planning. Not applicable to individual patient-level clinical decision-making.

Abstract

COVID-19 is a highly contagious disease transmitted primarily through human contact. Therefore, understanding population mobility is essential for predicting COVID-19 case trends. In this paper, we propose a novel deep learning approach for forecasting new COVID-19 cases using a neural architecture called Neural Basis Expansion Analysis for Interpretable Time Series (N-BEATS). The N-BEATS model effectively handles long input sequences and large output horizons without information loss or increased computational complexity. We compare the performance of N-BEATS with a state-of-the-art benchmark model, LSTM-Markov, across four major countries: the United States, the United Kingdom, Russia, and Brazil. Three distinct COVID-19 datasets from Google, Apple, and Our World in Data (OWID) were used in this study. Incorporating Google and Apple mobility data as covariates enhances both the accuracy and interpretability of the N-BEATS model. Our results show that N-BEATS consistently outperforms LSTM-Markov across all datasets and countries, consistently yielding lower Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). Furthermore, the N-BEATS model with covariates outperforms its counterpart without covariates, indicating that mobility data provide substantial value for forecasting new COVID-19 cases. Overall, this study demonstrates the effectiveness of the N-BEATS architecture in capturing pandemic dynamics and offers valuable insights for policymakers and public health officials in managing future outbreaks.

References

1. 1.Nazir, A., Shorfuzzaman, M., Lotfi, M. L., Kamalov, F., Badawi, S., Takruri, M., & Jallad, A.-H. (2026). Forecasting COVID-19 new cases using NBEATS deep learning and mobility data. PLOS ONE. https://doi.org/10.1371/journal.pone.0350264