Infectious disease prediction with kernel conditional density estimation.

Area of Science:

• Epidemiology
• Biostatistics
• Computational Biology

Background:

• Accurate prediction of infectious disease incidence is crucial for public health.
• Existing statistical models face challenges in capturing complex disease dynamics.

Purpose of the Study:

• To develop and evaluate a novel statistical approach for predicting infectious disease incidence.
• To improve forecasting accuracy for individual weeks and peak disease periods.

Main Methods:

• Kernel conditional density estimation (KCDE) with novel periodic and fully parameterized bandwidth matrix components.
• Copulas to link weekly incidence predictions into joint distributions.
• Application to dengue fever and influenza data, compared against baseline models.

Main Results:

• KCDE demonstrated superior performance for predicting weekly dengue incidence.
• KCDE showed more consistent predictions for peak incidence timing compared to baseline models.
• The periodic kernel component significantly improved incidence prediction accuracy.

Conclusions:

• The developed KCDE approach offers improved infectious disease incidence prediction capabilities.
• This method is particularly beneficial for diseases with heterogeneous seasonal patterns, like dengue fever.
• Enhanced predictive models can better support public health decision-making.