The increasing consumption of pellets has recently driven stove manufacturers to design more efficient combustion devices. However, due to the inherent variability of biomass, the expected performance is often unmet. Therefore, we propose a surrogate model-based approach to optimize combustion for pellet stoves. To do so, we first analysed the data via Analysis of Variance (ANOVA, feature selection) and t-distributed Stochastic Neighbor Embedding (t-SNE, data visualization). Subsequently, a Gaussian Process Regression (GPR) was implemented to predict Carbon Monoxide (CO) emissions across different stove settings, leveraging its strong predictive capabilities with small datasets. Finally, performance metrics were employed to analyze the predictive accuracy of the model. Several experimental scenarios were tested. It was observed that the model is not yet able to generalize the results due to a lack of quantity and quality in the data used. However, the first results show great potential and will serve as a basis for optimizing the combustion of the pellet stove.