Prediction of Physical Quality Parameters of Frozen Shrimp (Litopenaeus vannamei): An Artificial Neural Networks and Genetic Algorithm Approach

• Published in: Food and bioprocess technology Vol. 7; no. 5; pp. 1433 - 1444
• Main Authors: Ahmad, Imran, Jeenanunta, Chawalit, Chanvarasuth, Pisit, Komolavanij, Somrote
• Format: Journal Article
• Language: English
• Published: Boston Springer-Verlag 01.05.2014; Springer US; Springer Nature B.V
• Subjects: Agriculture; Algorithms; Artificial neural networks; Back propagation; Biotechnology; Chemistry; Chemistry and Materials Science; Chemistry/Food Science; chewiness; cohesion; Color; correlation; Correlation coefficient; Correlation coefficients; Data points; Decapoda; Dependent variables; Food Science; Freezing; Frozen foods; Genetic algorithms; hardness; linear models; Litopenaeus vannamei; Mathematical models; Neural networks; Original Paper; Parameters; prediction; Predictions; Root-mean-square errors; shrimp; storage time; texture; Thawing; Frozen shrimp; Thawing rate; Genetic algorithm; Artificial neural networks; Freezing rate; Storage temperature; GANN
• ISSN: 1935-5130; 1935-5149
• DOI: 10.1007/s11947-013-1135-3

Application of genetic algorithm to optimize an artificial neural network (ANN) model for predicting end-of-storage quality parameters of frozen shrimp (Litopenaeus vannamei), which influence consumer purchase decisions, is demonstrated in this paper. Freezing rate (FR), thawing rate (TR), storage time, width, thickness, and length of frozen shrimp were measured and chosen as input variables to train the ANN against Commission International de l’ Eclairage Color L*a*b* values, and textural properties (hardness, cohesiveness, and chewiness) as dependent variables. Experimentally obtained randomized data points (500) were used to develop the network, of which 20 % were used for testing the network, as an unseen environment. The developed genetic algorithm–artificial neural network (GANN) which included one hidden layer with 3–17 neurons successfully predicted color and textural values with correlation coefficient, R ², of >0.9 and root mean square error (RMSE) of <1.6. The redness (a*) and cohesiveness took the longest training time and highest number of generations, as compared to the other parameters. Percent relative importance of input variables to output variables indicated that TR, FR, and storage time were the most important variables for the prediction of color and texture parameters. The results are compared with multiple linear regression (MLR) and ANN trained with backpropagation (BP) algorithm. The results indicate that the GANN model shows much better prediction, as compared to MLR and BP with smallest RMSE and highest R ².