From Policy to Prediction: Assessing Forecasting Accuracy in an Integrated Framework with Machine Learning and...

Area of Science:

• Epidemiology
• Computational Biology
• Machine Learning

Background:

• Accurate infectious disease forecasting is crucial for public health interventions.
• Previous hybrid models estimated transmission rates using machine learning (ML), but only tested Gradient Boosting Models (GBMs).
• The comparative performance of various ML models in this context remained unexplored.

Purpose of the Study:

• To compare the forecasting accuracy of Gradient Boosting Models (GBMs), linear regressions, k-nearest neighbors, and Bayesian networks (BNs).
• To evaluate these models in predicting COVID-19 infected cases in the US and Canadian provinces.
• To assess model performance based on policy indices for future disease spread.

Main Methods:

• Implemented and compared four ML models: GBMs, linear regressions, k-nearest neighbors, and BNs.
• Utilized policy indices to estimate disease transmission rates.
• Applied models to forecast COVID-19 cases in the United States and Canadian provinces over 35 days.
• Analyzed Mean Absolute Percentage Errors (MAPE) for model comparison.

Main Results:

• No significant difference in MAPE was found across ML models on the combined dataset.
• Bayesian networks (BNs) demonstrated superior performance in fitting most training datasets.
• While significant differences were noted in two provinces, posthoc tests found no significant pairwise differences.

Conclusions:

• Machine learning models exhibit comparable overall forecasting power for infectious diseases.
• Bayesian networks are particularly effective for data-fitting applications in epidemiological modeling.
• Further research can explore diverse ML architectures for enhanced disease spread prediction.