Multivariate and AI-based modeling of cold-pressing efficiency in oily and confectionery sunflower seed hybrids.

BACKGROUND: This study provides an in-depth evaluation of newly developed oilseed and confectionery sunflower hybrids through the characterization of seed morphology, physical properties, oil and moisture content, and mechanical strength. The seeds were cultivated in Serbia and Argentina over two seasons, and cold pressed under controlled conditions to evaluate oil yield, pressing capacity, and processing behavior. RESULTS: Significant differences were observed among hybrids and cultivation locations in terms of seed traits and oil content. Oilseed hybrids exhibited substantially higher oil extraction efficiency (41.63-75.61%) and pressing capacities (up to 30 kg h-1) compared to confectionery hybrids (20.10-48.40% yield; ~15 kg h-1 capacity), aligning with their higher seed oil content. An artificial neural network (ANN) model was developed using measured seed morphological, compositional and physical characteristics, seed and press cake moisture and oil content, and seed mass as input variables (18 inputs) to predict oil and seed yields, oil and seed flow rates, oil outlet temperature, pressing time, obtained oil mass, and press cake mass (eight outputs). The optimal MLP 18-11-8 model demonstrated excellent predictive ability, with R2 values of 0.970, 0.898, and 0.924 for training, testing, and validation datasets, respectively. Sensitivity analysis highlighted seed oil content and input mass as the most influential factors for oil yield and capacity, while residual oil in press cake negatively affected oil recovery. CONCLUSION: This integrative approach, combining comprehensive physical and mechanical seed evaluation with data-driven artificial neural network modeling, provides a novel framework for predicting and optimizing cold-pressing performance in sunflower hybrids. The findings offer valuable insights for breeding programs, industrial processing, and machine design tailored to specific hybrid profiles. © 2026 Society of Chemical Industry.