AI- and Machine Learning-Assisted Optimization of a Low-Sodium Functional Seasoning Powder Enriched with Mushroom and Seaweed Extracts
DOI:
https://doi.org/10.51483/IJAIML.2766Keywords:
Artificial intelligence; machine learning; low-sodium food; seasoning powder; mushroom extract; seaweed extract; potassium chloride; umami; antioxidant activity; functional food; formulation optimization.Abstract
Excessive dietary sodium intake is associated with an increased risk of hypertension and cardiovascular and renal disorders, creating a need for sodium-reduced food products that maintain acceptable sensory quality. The present study focused on the development of a low-sodium functional seasoning powder incorporating mushroom and seaweed extracts as natural flavour-enhancing and functional ingredients. Three formulations were prepared by replacing 25%, 50%, and 75% of sodium chloride (NaCl) with potassium chloride (KCl), while maintaining mushroom extract and seaweed extract at 5% and 3%, respectively. The formulations were evaluated for proximate composition, physicochemical characteristics, antioxidant activity, sensory properties, and short-term storage stability.
The formulation containing 50% replacement of NaCl with KCl (F2) was selected as the experimentally promising formulation based on the available formulation and sensory results. F2 contained 14.63% protein, 12.48% ash, 3.21% fat, 63.86% carbohydrate, and 5.82% moisture. The pH and bulk density were 6.42 and 0.54 g/mL, respectively. DPPH and ABTS radical-scavenging activities were 68.4% and 72.1%, respectively. Sensory scores for appearance, aroma, taste, umami intensity, and overall acceptability were 8.1, 7.9, 8.3, 8.5, and 8.2, respectively, on a 9-point hedonic scale. The formulation remained free-flowing with no major observable sensory deterioration during 30 days of ambient storage.
An artificial intelligence and machine learning (AI/ML) framework is proposed to extend the formulation study by predicting sodium content, antioxidant activity, umami intensity, and sensory acceptability from formulation and processing variables. Multiple linear regression, random forest regression, support vector regression, gradient boosting, and artificial neural networks are proposed as candidate models. Because the current experimental dataset is limited, predictive accuracy and model-performance values are not claimed. The proposed framework provides a basis for future data-driven, multi-objective optimization of reduced-sodium functional seasoning formulations.
