Predicting malaria outbreak in The Gambia using machine learning techniques

Area of Science:

• Epidemiology
• Machine Learning
• Environmental Science

Background:

• Malaria remains a significant global health threat, particularly in tropical regions.
• It impacts public health and economic development, necessitating effective control strategies.
• Predictive modeling can aid in proactive malaria outbreak management.

Purpose of the Study:

• To develop and compare machine learning models for predicting malaria outbreaks in The Gambia.
• To identify the most accurate algorithms for malaria incidence forecasting.
• To assess the impact of meteorological and non-climatic factors on prediction accuracy.

Main Methods:

• Evaluated eight machine learning algorithms: C5.0 decision trees, artificial neural networks, k-nearest neighbors, support vector machines (linear and radial kernels), logistic regression, extreme gradient boosting, and random forests.
• Utilized historical meteorological data for The Gambia.
• Employed 10-fold cross-validation, repeated five times for robust model validation.

Main Results:

• Extreme gradient boosting and decision trees demonstrated the highest prediction accuracy (93.3%) on the testing set.
• Random forests achieved a high accuracy of 91.5%.
• Support vector machine with a linear kernel showed lower accuracy (84.8%) and poorer specificity.

Conclusions:

• Machine learning models, particularly extreme gradient boosting and decision trees, are effective for predicting malaria outbreaks in The Gambia.
• The integration of climatic and non-climatic features is essential for accurate malaria outbreak prediction.
• Accurate forecasting can support targeted public health interventions and resource allocation for malaria control.