Skin contamination is a critical defect in aluminum alloy extrusion, leading to rejection of contaminated profiles due to poor mechanical properties. Experimental, empirical and numerical modeling approaches to identifying skin contamination in industry are either time-consuming, inaccurate or expensive. The present study integrates finite element method (FEM) simulations and machine learning (ML) to predict skin formation and evolution, offering a powerful tool for optimizing extrusion and advancing smart manufacturing practices. FEM simulations are calibrated and validated against several industrial case studies, after which artificial neural networks (ANNs) are trained with the simulation dataset, significantly reducing the computational cost of future predictions. The proposed approach not only enables accurate prediction of skin contamination but also provides insight into the die and process parameters most critical to skin defect formation.