Repository logo

Browsing by Author "Vlajnic, Marco M., author"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Novel assessments of country pandemic vulnerability based on non-pandemic predictors, pandemic predictors, and country primary and secondary vaccination inflection points
    (Colorado State University. Libraries, 2024) Vlajnic, Marco M., author; Simske, Steven, advisor; Cale, James, committee member; Conrad, Steven, committee member; Reisfeld, Bradley, committee member
    The devastating worldwide impact of the COVID-19 pandemic created a need to better understand the predictors of pandemic vulnerability and the effects of vaccination on case fatality rates in a pandemic setting at a country level. The non-pandemic predictors were assessed relative to COVID-19 case fatality rates in 26 countries and grouped into two novel public health indices. The predictors were analyzed and ranked utilizing machine learning methodologies (Random Forest Regressor and Extreme Gradient Boosting models, both with distribution lags, and a novel K-means-Coefficient of Variance sensitivity analysis approach and Ordinary Least Squares Multifactor Regression). Foundational time series forecasting models (ARIMA, Prophet, LSTM) and novel hybrid models (SARIMA-Bidirectional LSTM and SARIMA-Prophet-Bidirectional LSTM) were compared to determine the best performing and accurate model to forecast vaccination inflection points. XGBoost methodology demonstrated higher sensitivity and accuracy across all performance metrics relative to RFR, proving that cardiovascular death rate was the most dominant predictive feature for 46% of countries (Population Health Index), and hospital beds per thousand people for 46% of countries (Country Health Index). The novel K-means-COV sensitivity analysis approach performed with high accuracy and was successfully validated across all three methods, demonstrating that female smokers was the most common predictive feature across different analysis sets. The new model was also validated with the Calinski-Harabasz methodology. Every machine learning technique that was evaluated showed great predictive value and high accuracy. At a vaccination rate of 13.1%, the primary vaccination inflection point was achieved at 83.27 days. The secondary vaccination inflection point was reached at 339.31 days at the cumulative vaccination rate of 67.8%. All assessed machine and deep learning methodologies performed with high accuracy relative to COVID-19 historical data, demonstrated strong forecasting value, and were validated by anomaly and volatility detection analyses. The novel triple hybrid model performed the best and had the highest accuracy across all performance metrics. To be better prepared for future pandemics, countries should utilize sophisticated machine and deep learning methodologies and prioritize the health of elderly, frail and patients with comorbidities.