This thesis explores the improvement of building energy efficiency in France to reduce greenhouse gas emissions. The main objective is to develop an explainable predictive model based on EPCs (Energy Performance Certificates), despite uncertain and incomplete data. The first chapter describes the data fusion process to create a unified database. The second proposes a geostatistical regression model (Mixture Kriging), which performs well at the local scale but is limited nationally due to its complexity. The third introduces Joint Kriging, a fuzzy classification model that is effective on a large scale. Finally, the fourth compares different approaches, including Random Forest and FKNN, to improve prediction. This work contributes to environmental science and applied mathematics, helping policymakers identify energy-inefficient buildings and achieve sustainability goals.

The jury is composed of:

• Elena DI BERNARDINO, Professor, Université Côte d’Azur, Nice, Reviewer
• Robin GIRARD, Professor, Research Director, Mines Paris, Reviewer
• Natacha GONDRAN, Professor, Mines Saint-Etienne, Examiner
• Gaël POETTE, Research Engineer, CEA/Professor, Bordeaux INP, Examiner
• Simon ROUCHIER, Associate Professor, Polytech Annecy Chambéry, Examiner
• Didier RULLIERE, Professor, Mines Saint-Etienne, Thesis Director
• Jonathan VILLOT, Assistant Professor, Mines Saint-Etienne, Co-supervisor.