Investigation of the influence of seed size on the biochemical and physical characteristics of black cumin (Nigella sativa) with machine learning techniques

This research aimed to examine the impact of black cumin (Nigella sativa) seed sizes (S1: >1.6 mm, S2: 1.4-1.6 mm, S3: 1.2-1.4 mm, S4: <1.2 mm) on various physical and chemical properties (crude oil, ash, fiber, carbohydrate, protein, and volatile oil content) and to enhance seed classification utilizing machine learning (ML) algorithms. Ten different ML algorithms (LR: Logistic Regression, SMO: Sequential Minimal Optimization, J48: Decision Tree, LMT: Logistic Model Tree, RT: Random Tree, RF: Random Forest, SL: Simple Logistic, MLP: Multi-Layer Perceptron KNN: k-nearest neighbor, KStar: instance-based classifier) were used to discriminate between seed classes. The MLP model performed the best with 96.25% accuracy. According to the results of analysis of variance and multivariate analyses (biplot and canonical correlation), seed size significantly affected seed physical and chemical characteristics. Analysis of variance (ANOVA) and multivariate methods (biplot and canonical correlation) showed significant differences (p < 0.01) in physical parameters, such as mass (0.0018-0.0026 g), thousand kernel weight (1.97-2.51 g), and geometric mean diameter (1.49-1.68 mm), between seed size groups. Chemically, smaller seeds (S3, S4) had higher levels of crude oil (31.62-32.77%) and linoleic acid (up to 62.10%), while larger seeds (S1, S2) had higher levels of fiber (6.62-6.97%), ash (4.63-4.88%), and carbohydrates (31.68-33.91%). Seed size S2 had the highest volatile oil content (0.63%) and thymoquinone levels (11.75%), underscoring its exceptional potential for medicinal, industrial, and direct food applications. Based on these results, it is advisable to utilize black cumin seeds for various applications in the food, cosmetic, and pharmaceutical industries by assessing different seed sizes.