Spatio-temporal data mining to study the risk of malaria residual transmission at a landscape scale in rural West Africa

Abstract

The fight against malaria transmission is currently stalling. To reinvigorate progress, we must shift from a universal approach of prevention to a targeted one, adapted to the local transmission risk profile. Such strategy requires to characterise, understand, and predict the risk of transmission of malaria at fine spatial and temporal scales, i.e. scales suitable for local decision-making. In this thesis, we have attempted to explain and evaluate the spatial and temporal predictability of several entomological indicators of transmission risk at a landscape scale in rural West Africa (in Burkina Faso and Côte d'Ivoire) : presence, abundance of anopheles, physiological and behavioral resistance to insecticides. We used heterogeneous, spatio-temporal, multi-source and multi-scale entomological and environmental data, and data mining methods (among others, based on interpretable machine learning techniques), in a holistic-inductive approach to scientific knowledge generation. Our results showed strong spatio-temporal heterogeneities in vector abundances at the village scale, and relative homogeneities in the prevalence of vector resistances. Based on the associations captured by the statistical models, we made numerous hypotheses on the environmental determinants (climatic, landscape, socio-cultural, etc.) of the various entomological indicators studied ; in other words, on the impact of environmental conditions on the vectors' life traits. Our models were able to acurately forecast vector abundances at the village scale several weeks ahead, which was not the case for the prevalence of insecticide resistance. At the end of this work, we make proposals for the improvement of (i) current vector control methods, (ii) the use of (geo)data science and data engineering in general, and statistical modelling in particular, for malaria research and control, and (iii) tools for the surveillance and prevention of malaria transmission risk at the loca l scale in rur