A Comparative Analysis of Predictive Approaches for COVID-19 RecoveryRates: Evaluating Regional Variations and Uncertainty
Precision Journal of Applied Mathematics and Statistics · 2026
DOI: 10.67887/pd2601208006 ↗Kurz gefasst
Fachartikel in der Zeitschrift Precision Journal of Applied Mathematics and Statistics, 2026. Verfasst von Ayesha Sultan. Die Arbeit nennt 65 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Introduction
Accurate forecasting of COVID-19 recovery rates is important for understanding the development of regional recovery patterns. Predictive models should, however, be evaluated not only according to point-prediction accuracy but also in relation to temporal learning, uncertainty representation, interpretability, and possible failure under limited-information conditions. Methodology: This study compares five predictive approaches: Linear Regression, Random Forest, Bayesian Linear Regression, Autoregressive Integrated Moving Average and Long Short-Term Memory. The source dataset contained 156,292 daily observations from 22 January to 15 November 2020. To maintain consistency with the study’s temporal figures, the empirical analysis was restricted to 22 January–31 October 2020. Following aggregation, removal of zero-denominator observations, and exclusion of seven invalid recovery-rate values above one, the final dataset contained 46,020 country–date observations representing 225 Country/Region units. The earliest 227 dates, comprising 35,167 observations, were used for training, while the most recent 57 dates, comprising 10,853 observations, formed the test set. Region_ID was the sole predictor used by the three tabular models, while ARIMA and LSTM modelled chronologically ordered recovery-rate observations. Healthcare capacity, population density, government interventions, and socioeconomic conditions were discussed conceptually but were intentionally excluded from empirical modelling. Model performance was evaluated using Root Mean Squared Error and Mean Absolute Error.
Results
LSTM achieved the lowest overall error, with an RMSE of 0.292 and an MAE of 0.239. Random Forest was the strongest tabular model, with an RMSE of 0.2977 and an MAE of 0.2463. Bayesian Linear Regression produced higher point-prediction errors but provided explicit probabilistic uncertainty estimates.
Conclusion
The findings demonstrate that model selection should consider prediction error, temporal structure, uncertainty, and interpretability rather than accuracy alone. They also show that performance obtained from a restricted feature space should not be interpreted as evidence of causal regional determinants.
Zitieren
APA
Sultan, A. (2026). A Comparative Analysis of Predictive Approaches for COVID-19 RecoveryRates: Evaluating Regional Variations and Uncertainty. Precision Journal of Applied Mathematics and Statistics. https://doi.org/10.67887/pd2601208006
Vancouver
Sultan A. A Comparative Analysis of Predictive Approaches for COVID-19 RecoveryRates: Evaluating Regional Variations and Uncertainty. Precision Journal of Applied Mathematics and Statistics. 2026. doi:10.67887/pd2601208006
BibTeX
@article{Sultan2026,
author = {Sultan, Ayesha},
title = {A Comparative Analysis of Predictive Approaches for COVID-19 RecoveryRates: Evaluating Regional Variations and Uncertainty},
journal = {Precision Journal of Applied Mathematics and Statistics},
year = {2026},
doi = {10.67887/pd2601208006}
}
EUVIMED