Predicting dengue importation into Europe, using machine learning and model-agnostic methods.

Area of Science:

• Epidemiology and Public Health
• Computational Biology and Bioinformatics
• Network Science

Background:

• Geographical spread of dengue is a significant global public health issue.
• Air transport networks facilitate dengue importation from endemic to non-endemic regions.
• Predicting dengue importation is complex due to dynamic and heterogeneous air travel patterns.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) algorithms in predicting dengue importation.
• To identify key predictors influencing dengue importation events.
• To explore the interpretability of ML models for understanding prediction drivers.

Main Methods:

• Trained four ML classifiers using 6 years of historical dengue importation data for 21 European countries.
• Incorporated air transport network connectivity indices and centrality measures as predictors.
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AUC), sensitivity, and specificity.
• Applied model-agnostic methods for prediction explanation.

Main Results:

• The best performing ML model achieved a high AUC of 0.94 and a maximized sensitivity of 0.88.
• Most important predictors identified: source country's dengue incidence rate, population size, and air passenger volume.
• Network centrality measures also demonstrated influence on prediction accuracy.

Conclusions:

• Machine learning models exhibit high predictive performance for dengue importation.
• Model-agnostic interpretability methods provide valuable insights into prediction factors.
• This approach can inform the development of operational early warning surveillance systems for dengue.