Main Article Content



The objective of this study was to apply artificial  neural network (ANN) to enable accurate and fast prediction of moisture, protein. lysine and methionine contents of ftishmeal. The several wevelenqths of near intrared absorbance, range from 900 to 2,000 nm, were selected for training and validating ANN on each chemical component by stepwise multiple linear regression analysis. Tne ANN with three, five. seven and nine nodes at hidden layer were trained using 35 samples for moisture and protein, 33 samples for lysine and 30 samples for metionine Validating was conducted on 10 independent samples. The results of validating indicated that the best of protoin prediction was achieved by ANN with seven nodes at hidden layer for moisture. five nodes for protein and methionine, and three nodes for lysine. The standard error of prediction, coefficient of variation and ratio of standard deviation and standard error of prediction respoctivety were 0.61%, 4.81%, and 6.89 for moisture contents; 2.99%, 6.43% and 3.34 for protein contents: 0.14%, 11.32% and 3.04 for lysine contents: and 0.07%. 10.50% and 2.16 for methionine contents. With the same data entry. the ANN could predict with better performance than did by multiple linear regression

Keywords: artificial neural network, near infrared, fishmeal

Diterima: 25 September 2007, Disetujui: 9 Nopember 2007














Article Details