Exploring predictive frameworks for malaria in Burundi.

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• Malaria infection rates are increasing in Burundi despite existing health service access and intervention programs.
• Heterogeneous data offers potential for developing predictive models for malaria cases.

Purpose of the Study:

• To develop and compare predictive frameworks, specifically Generalized Linear Models (GLM) and Artificial Neural Networks (ANN), for forecasting malaria cases in Burundi.
• To identify key demographic and environmental factors influencing malaria transmission patterns.

Main Methods:

• Utilized heterogeneous data to construct predictive models, including GLM and ANN.
• Analyzed malaria case data across four sub-groups and the general population.
• Evaluated model performance for accuracy in prediction.

Main Results:

• Over half of malaria infections were concentrated in pregnant women and children under 5 years, particularly those aged 12-59 months.
• The ANN model demonstrated superior performance in predicting total malaria cases compared to the GLM.
• Identified that higher education rates and increased use of Insecticide Treated Bed Nets (ITNs) correlated with decreased malaria cases, while other factors showed an increasing effect.

Conclusions:

• ANN models provide a more effective approach for malaria case prediction in Burundi.
• Malaria control strategies should emphasize the distribution of ITNs and community-wide sensitization, especially in densely populated areas.
• Timely prediction of malaria cases is crucial for proactive intervention, epidemic mitigation, and reducing socioeconomic impact.