Artificial intelligence predictability of moisture, fats and fatty acids composition of fish using low frequency Nuclear Magnetic Resonance (LF-NMR) relaxation

Abstract

Moisture, fats and fatty acids of 14 pelagic and demersal fishes were measured by conventional chemical analysis to relate these with the proton relaxation using Low Frequency Nuclear Magnetic Resonance (LF-NMR). Artificial intelligence was used to assess the predictability of composition using six relaxation parameters of LF-NMR. Multiple linear regression showed significant prediction for moisture (W) (P < 0.00001), total fat (F) (P < 0.0001), ω-6 fatty acid (O6) (P < 0.001), saturated fats (SF), fatty acids (FA), mono-unsaturated fatty acids (MU) and ω-3 fatty acid (O3) (P < 0.01). However, the highest regression coefficient was observed for water (R²: 0.490) and the lowest was observed for SF (R²: 0.224). The low regression coefficients indicated strong non-linear relationships exited between LF-NMR parameters and composition. However, decision tree showed higher regression coefficients for all compositions considered in this study (R²:0.780–0.694). In addition, it provided simple decision rules for the prediction of composition. General Regression Neural Network provided the highest prediction capability (R²:0.847–1.000 for training and 0.506–0.924 for validation).