Predicting local malaria exposure using a Lasso-based two-level cross validation algorithm.

Area of Science:

• Environmental Epidemiology
• Vector-Borne Disease Modeling
• Statistical Learning in Public Health

Background:

• Local environmental factors significantly influence malaria transmission dynamics.
• Understanding fine-scale heterogeneity in vector exposure is key for effective control strategies.

Purpose of the Study:

• To develop and compare predictive models for individual exposure to malaria vectors.
• To evaluate the performance of LASSO-based algorithms against traditional Generalized Linear Models (GLM) with backward selection.

Main Methods:

• Utilized entomological and environmental data from a cohort study in Benin.
• Compared a classical GLM with backward selection to various automatic LASSO-based algorithms.
• Employed a 2-level cross-validation strategy for model evaluation.

Main Results:

• The LASSO-based model demonstrated superior predictive power for space and time-dependent individual exposure to malaria vectors.
• While GLM could perform well with specific feature engineering, LASSO offered a more robust predictive approach.
• The study identified key environmental and entomological predictors of vector exposure.

Conclusions:

• LASSO-based algorithms offer a powerful tool for predicting individual malaria vector exposure.
• This predictive modeling approach can be adapted to various health science domains facing similar prediction challenges.
• Accurate prediction of vector exposure is vital for targeted malaria interventions.