A Dirichlet process model for classifying and forecasting epidemic curves.

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Forecasting future events, especially epidemic dynamics, is complex due to numerous influencing factors.
• Accurate epidemic forecasting is crucial for timely public health interventions.
• This study introduces a novel Dirichlet process (DP) model for influenza epidemic curve classification and forecasting.

Purpose of the Study:

• To develop and validate a Dirichlet process (DP) model for classifying and forecasting influenza epidemic curves.
• To compare the performance of the DP model against the Random Forest (RF) technique.
• To assess the DP model's ability to identify novel epidemic patterns and predict peak times.

Main Methods:

• A nonparametric Bayesian Dirichlet process (DP) model was employed.
• The DP model was validated using simulated influenza epidemics from an individual-based model.
• The model was applied to historical influenza-like illness (ILI) data from the CDC (1997-2013).
• Performance was compared against the Random Forest (RF) classification technique.

Main Results:

• The DP model demonstrated comparable performance to RF in identifying simulated epidemics.
• The DP model accurately forecasted epidemic peak times several days in advance for most simulations.
• The accuracy of identifying novel epidemic curves improved with increased data.
• Both methods showed higher accuracy in classifying epidemics with higher reproduction numbers (R).

Conclusions:

• The DP model offers unsupervised, semi-supervised, or fully supervised learning capabilities, unlike the fully supervised RF.
• While RF is computationally faster, the DP model's flexibility allows it to classify previously unobserved epidemic curves.
• A combined approach utilizing both DP and RF models may offer synergistic benefits for epidemic forecasting.