Conducted jointly by UAC and UP1, this research explores the modeling of malaria-related morbidity in Tori-Bossito.
The Essentials: Conducted jointly by UAC and UP1, this research explores the modeling of malaria-related morbidity in Tori-Bossito.
Bienvenue Kouwaye, researcher at SAMM (Université d’Abomey-Calavi = University of Abomey Calavi [Cotonou, Bénin]).
Thesis defended in 2018.
This research was carried out under a Franco-Beninese joint supervision (cotutelle), ensuring simultaneous grounding in local field realities and international academic standards.
Context and Research Question
Malaria remains a major public health concern, particularly in West Africa, where it continues to generate high rates of morbidity and mortality. The Tori-Bossito region of Benin has varied environmental characteristics that influence Anopheles mosquito density, the vector of the disease. Understanding the environmental factors that affect this density is essential for predicting malaria exposure risk and guiding public health interventions.
The research conducted by Bienvenue Kouwaye focuses on identifying the relevant environmental variables for modeling malaria exposure risk. It approaches this problem from the angle of variable selection and prediction, using modern statistical approaches. The main objective is to determine an optimal subset of variables that can explain variability in Anopheles density and effectively predict risk at the village and household level.
Methodology
The thesis adopts a two-part methodological approach to address the challenges posed by malaria risk modeling. In the first part, a method based on generalized linear mixed models (GLMM) is implemented. This approach incorporates backward-type variable selection and accounts for random effects to reflect the hierarchical structure of the data. This method makes it possible to identify an optimal subset of variables relevant to predicting malaria exposure risk.
However, the limitations of this method, notably convergence issues and correlations between data, led to the exploration of machine learning approaches. In the second part, a method combining the generalized linear model (GLM), Lasso, and a two-level stratified cross-validation is proposed. This method automatically generates interactions between variables and performs a more robust variable selection. The variables selected are then debiased by the GLM to improve prediction quality.
Key Findings
The results of this research indicate that significant interactions exist between several environmental variables and malaria risk. The factors identified as most determinant include season, average rainfall, population density and vegetation. Applying variable selection algorithms made it possible to overcome the limitations of expert pre-processing, leading to a notable improvement in prediction quality.
The integration of machine learning methods also made it possible to optimize the sparsity of the selected variable subset. The results show a decrease in CPU computation time, reinforcing the efficiency of the statistical methods applied. Moreover, the two-level stratified cross-validation ensured the robustness of predictions in epidemiology.
Discussion and Outlook
Bienvenue Kouwaye's thesis highlights the importance of statistical learning in the fight against malaria, providing methodological tools adapted to specific contexts such as Tori-Bossito. By identifying interactions between environmental variables and malaria risk, this research paves the way for targeted interventions grounded in empirical data.
The results obtained underscore the need to integrate these models into existing public health strategies. Applying advanced modeling techniques could also be extended to other at-risk regions, thereby strengthening malaria prevention and control efforts. Collaboration between researchers and policymakers is essential to translate these results into concrete action on the ground, improving the capacity to respond to this persistent disease.
The next steps in this research could explore the integration of other types of data, such as those related to human behavior or other environmental factors, to further enrich predictive models. These advances could contribute to a better understanding of malaria dynamics and to more targeted and effective interventions.
Accéder à l'étude complète
Sources and Access
Bienvenue Kouwaye. Contributions de l'apprentissage statistique aux méthodes GLMM et LASSO: Application à la modélisation statistique de la morbidité liée au paludisme à Tori-Bossito (Bénin). Statistiques [math.ST]. Université d'Abomey-Calavi (Bénin), 2018. Français. ⟨NNT : ⟩. ⟨tel-01736933⟩