In organic synthesis, many molecular and macroscopic parameters can influence the outcome of chemical reactions. It is therefore difficult to correlate the obtained yields with the reaction conditions. This project aims to develop interpretable machine learning models to predict and improve the efficiency of oxidation reactions of electron-deficient heterocycles, a real challenge in organic chemistry. The main challenge will be to best represent and leverage the variables associated with the complexity of a real reaction system (chemical nature of the substrate, temperature, reaction time, etc.) to feed machine learning algorithms and extract clear rules. The ultimate goal is to provide chemists with predictive tools to rationalize and develop these transformations.