Time series forecasting for tuberculosis incidence employing neural network models

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Tuberculosis (TB) control is a global health priority.
• Effective national programs require accurate forecasting of disease incidence.
• Computational tools can aid in developing strategies for patient management and treatment.

Purpose of the Study:

• To evaluate the performance of different artificial neural network models for Tuberculosis time series forecasting.
• To compare traditional computational methods with connectionist approaches for TB case prediction.
• To identify the most effective forecasting model for supporting public health strategies in Colombia.

Main Methods:

• Utilized time series forecasting with artificial neural networks.
• Trained models using reported Tuberculosis case data from Colombia's national vigilance institution.
• Proposed and compared three neural models: nonlinear autoregressive, recurrent neural network, and radial basis functions.
• Evaluated model performance using the mean average percentage error (MAPE).

Main Results:

• Models based on traditional methods demonstrated superior forecasting performance compared to connectionist models.
• The nonlinear autoregressive and radial basis function models showed better accuracy in predicting Tuberculosis incidence.
• The recurrent neural network model exhibited lower performance in this specific forecasting task.

Conclusions:

• Traditional computational models are more effective for Tuberculosis time series forecasting in the Colombian context.
• Accurate incidence forecasting provides dynamic information crucial for health authorities.
• These findings support the development of enhanced strategies for Tuberculosis control and disease management.