<p>Biomass is a valuable renewable resource to produce sustainable fuels and chemicals; its efficient utilization depends on accurate compositional characterization. This study presents a machine learning framework to predict the elemental composition of biomass carbon, hydrogen, oxygen, nitrogen, and sulfur from proximate analysis, net calorific value, and biomass clustering. A dataset of 426 samples from the Phyllis2 database was used, with outliers removed using the Local Outlier Factor (LOF) method. Biomass samples were first grouped using k-means clustering based on proximate analysis and net calorific value, identifying four distinct clusters with different compositional characteristics. These clusters were encoded and, together with the input variables, used as inputs to an artificial neural network (ANN) to predict elemental composition. The ANN achieved high predictive performance, with a mean squared error of 1.84 and R² of 0.994 on the test set. For validation, new samples were classified using a logistic regression model trained on the clustering results, and their elemental composition was subsequently predicted using the ANN. The proposed approach outperformed literature models for carbon, hydrogen, and oxygen. Lower accuracy for nitrogen and sulfur was attributed to their high variability and skewed distributions. The integration of biomass classification extends the applicability of the model to a wide range of carbonaceous materials, including organic residues and coal, providing a practical tool for biomass selection in thermochemical processes such as gasification.</p> Graphical Abstract <p></p>