Rodrigue is a Doctor in Computer Science (Data Science). His thesis is entitled “Deep Learning on Attributed Graphs for Mapping Leptospirosis Risk”. He defended his PhD in August 2025 at the University of New Caledonia.

This thesis explored supervised learning methods applied to the risk mapping of leptospirosis in New Caledonia. To do so, a holistic approach is considered, by collecting, pre-processing and integrating a broad spectrum of variables, whether they are meteorological, environmental or socio-demographical. The risk mapping has been conducted according the leptospirosis surveillance data from 2011 to 2022, on a spatial scale finer than the municipal level, and on a monthly time step. This spatio-temporal granularity therefore posed a real challenge of imbalanced data. Combined with several sampling strategies, two approaches were developed.

The first approach involved an ensemble learning, coupled with under-sampling and hybrid sampling strategies for model training, as well as weighted prediction to optimize the approach’s effectiveness. Given conclusive results, an explainability component was developed, enabling the identification of most contributing factors to the leptospirosis risk. Despite satisfying results, this first approach required a large number of supervised learning models.

In parallel, this thesis investigates graph neural network (GNN) methods, focusing specifically on the optimal reduction of attributed graphs within a GNN model by integrating existing pooling techniques from the literature. The resulting hybrid method, named SpaPool, performed comparably to established approaches and demonstrated superior results on datasets consisting of small attributed graphs. Given the effectiveness of graph-based representations, the second approach to map leptospirosis risk using a single GNN model was developed. Combined with various sampling strategies, this method produced promising outcomes, yielding more balanced sensitivity and specificity scores compared to the ensemble approach.

The contributions presented in the thesis, both methodological and applied, offer new opportunities for studying other health-related and anthropogenic phenomena.

This thesis was under supervision of Prof. Nazha Selmaoui-Folcher, Full Professor in Computer Science at the University of New Caledonia and Prof. Philippe Fournier-Viger, Distinguished Professor in Computer Science at the Shenzhen University (China).

During this thesis, Rodrigue has given classes to Bachelor students. His main courses involved Advanced programming in Python, Graph Theory, and Database management.