[Early warning on measles through the neural networks]

Area of Science:

• Epidemiology
• Computational Science
• Public Health

Background:

• Measles surveillance is crucial for timely intervention.
• Predictive modeling can improve early warning systems for infectious diseases.
• Analyzing historical measles data can reveal patterns for forecasting.

Purpose of the Study:

• To evaluate the efficacy of neural networks for measles early warning.
• To compare the performance of different neural network models in predicting measles prevalence and incidence.
• To assess the feasibility of using neural networks in public health surveillance systems.

Main Methods:

• Utilized monthly and weekly measles surveillance data from Wuhan city (1986-2006).
• Developed a dynamic time series model using two-layer back propagation (BP) neural networks.
• Compared BP networks with probabilistic neural networks (PNN) for forecasting tasks.

Main Results:

• A two-layer BP network achieved a correlation coefficient of 0.85 with acceptable convergence speed.
• BP networks were more suitable for monthly value forecasting, while PNN excelled at weekly classification.
• The choice of network (BP vs. PNN) depended on data availability, with BP preferred for larger datasets.

Conclusions:

• Neural networks, particularly BP networks for ample data and PNN for sparse data, offer a feasible approach for measles early warning systems.
• This predictive modeling strategy can enhance the timeliness and accuracy of public health responses to measles outbreaks.
• The study demonstrates the potential of computational methods in strengthening infectious disease surveillance.