Machine Learning Forecast of Growth in COVID-19 Confirmed Infection Cases with Non-Pharmaceutical Interventions and Cultural Dimensions: Algorithm Development and Validation

Arnold Ys Yeung, François Roewer-Després, Laura C. Rosella, Frank RudziczmedRxiv 2021 · DOI

Abstract

National governments have implemented non-pharmaceutical interventions (NPIs) to control and mitigate against the COVID-19 pandemic. A deep understanding of these interventions is required. We investigate the prediction of future daily national Confirmed Infection Growths (CIG) – the percentage change in total cumulative cases across 14 days – using metrics representative of NPIs and cultural dimensions of each country. We combine the OxCGRT dataset, Hofstede’s cultural dimensions, and COVID-19 daily reported infection case numbers to train and evaluate five non-time series machine learning models. Our results demonstrate high $R^2$ values for in-distribution, out-of-distribution, and country-based cross-validation methods (0.959, 0.513, and 0.574 respectively) using random forest and AdaBoost regression. This work provides new considerations in using machine learning techniques with NPI and cultural dimensions data for predicting the national growth of confirmed infections.

Introduction

Predicting the trajectory of a pandemic is notoriously difficult. While standard epidemiological models (like the SIR model) are useful, they often rely on rigid assumptions and struggle to account for the human element: how people actually respond to government mandates. Two countries might implement the exact same "lockdown" policy, yet see vastly different results. This discrepancy is often driven by the interaction between Non-Pharmaceutical Interventions (NPIs) and underlying "Cultural Dimensions." This research moves beyond simple curve-fitting to investigate whether machine learning can leverage policy and culture to forecast Confirmed Infection Growth (CIG)—the 14-day percentage change in cumulative cases—across 114 different countries.

Methods and Datasets

The researchers integrated three primary data sources covering 114 countries from April 1 to September 30, 2020:

  • OxCGRT (Oxford COVID-19 Government Response Tracker): Daily metrics on 17 NPI categories, such as school closures and travel restrictions.
  • Hofstede’s Cultural Dimensions: Six metrics (e.g., Power Distance, Individualism) describing national cultural values.
  • JHU CSSE: Daily reported COVID-19 infection numbers used to calculate the CIG.

The study opted for non-time series machine learning models—Ridge Regression, Decision Tree, Random Forest, AdaBoost, and Support Vector Regression—to minimize sensitivity to noisy early-outbreak data. Robustness was ensured through three validation strategies: in-distribution (interpolation), out-of-distribution (forecasting), and country-based cross-validation (generalization).

Key Insights: The Power of Culture

A significant finding of this study is that a population's underlying values are often as predictive as the laws passed by their government. Using Mutual Information (MI) scores to rank feature importance, the researchers found that cultural dimensions frequently outperformed specific NPIs.

Mutual Information scores for Non-Pharmaceutical Intervention (NPI) features.

For instance, "Long-term orientation" achieved an MI score of 0.461, which was significantly higher than many specific NPIs like "International travel controls" (0.099). This suggests that cultural norms—such as individualism versus collectivism—are critical predictors of how effectively a country controls viral spread.

Mutual Information scores for Hofstede’s Cultural Dimensions.

Results and Performance

The models showed strong predictive power, particularly Random Forest and AdaBoost:

  • In-distribution: Achieved a very high $R^2$ of 0.959 (MAE 0.031).
  • Out-of-distribution (Forecasting): The $R^2$ dropped to 0.513 (MAE 0.089), reflecting the difficulty of long-term forecasting as conditions like variants and policy fatigue evolve.
  • Country-based (Generalization): Achieved an $R^2$ of 0.574 (MAE 0.167), suggesting the models can reasonably predict growth in a "new" country based on its culture and policies.

Discussion and Limitations

While powerful, the models face challenges. NPI enforcement varies significantly by region, and cultural scores for some areas were approximated. Furthermore, the 14-day prediction window cannot account for policy changes occurring within that timeframe. Finally, "confirmed cases" serve as a proxy for actual infections but are heavily influenced by national testing capacities. Future work could improve these forecasts by incorporating more granular, real-time data and refining cultural metrics to be more country-specific.

Summary updated

Abstract

Abstract Background National governments have implemented non-pharmaceutical interventions to control and mitigate against the COVID-19 pandemic. A deep understanding of these interventions is required. Objective We investigate the prediction of future daily national Confirmed Infection Growths – the percentage change in total cumulative cases across 14 days – using metrics representative of non-pharmaceutical interventions and cultural dimensions of each country. Methods We combine the OxCGRT dataset, Hofstede’s cultural dimensions, and COVID-19 daily reported infection case numbers to train and evaluate five non-time series machine learning models in predicting Confirmed Infection Growth. We use three validation methods – in-distribution, out-of-distribution , and country-based cross-validation – for evaluation, each applicable to a different use case of the models. Results Our results demonstrate high R 2 values between the labels and predictions for the in-distribution, out-of-distribution , and country-based cross-validation methods (0.959, 0.513, and 0.574 respectively) using random forest and AdaBoost regression. While these models may be used to predict the Confirmed Infection Growth, the differing accuracies obtained from the three tasks suggest a strong influence of the use case. Conclusions This work provides new considerations in using machine learning techniques with non-pharmaceutical interventions and cultural dimensions data for predicting the national growth of confirmed infections of COVID-19.
BibTeX
@misc{yeung2021machine,
  author = {Arnold YS Yeung and François Roewer-Després and Laura C. Rosella and Frank Rudzicz},
  title = {{Machine Learning Forecast of Growth in COVID-19 Confirmed Infection Cases with Non-Pharmaceutical Interventions and Cultural Dimensions: Algorithm Development and Validation}},
  year = {2021},
  doi = {10.1101/2021.01.04.21249235},
  howpublished = {medRxiv},
}
Figure