NOVA Through Two Lenses: From Nutrients to Food Processing Classification in Finland's National Food Database

FoodProX trained on 182,344 Open Food Facts products assigns NOVA groups in Finland’s Fineli with 55.9% of 4,092 scored items labeled NOVA 4.

• Why it matters: NOVA labels are inconsistent and often missing from national food composition databases, limiting ultra-processed food research and processing-intensity comparisons across settings.
• What they did: The nutrient-based FoodProX model predicts NOVA groups and a continuous FPro score from nutrient composition, evaluated on held-out Open Food Facts (45,526) and US FNDDS (2,971) before applying to Fineli.
• The result: Four-class accuracy was 0.855 in held-out Open Food Facts and 0.763 in FNDDS, with errors concentrated at the NOVA 3/4 boundary; UPF detection precision-recall was 0.940 and 0.964, and FPro ranking matched US FoodProX (Spearman rho = 0.81 in FNDDS, 0.79 in Fineli).