comparison of multiple linear regression and artificial neural network models to estimate of amino acid values in pearl millet hybrid based on chemical composition

pearl millet has tolerance to harsh growing conditions such as drought. it is at least equivalent to maize and generally superior to sorghum in protein content and metabolizable energy levels. thus it is of importance for poultry feeding. amino acid (aa) determination is expensive and time consuming. therefore nutritionists have prompted a search for alternatives to estimate aa levels. traditionally, two methods of predicting aa levels have been developed using multiple linear regression (mlr) with an input of either cp or proximate analysis. artificial neural networks (ann) may be more effective to predict aa concentration in feedstuff. therefore a study was conducted to predict the aas level in pearl millet with either mlr or ann. fifty two samples of pearl millet’s data lines contained chemical compositions and aas which collected from literature were used to find the relationship between chemical analysis as xi and aa contents as y. for both mlr and ann models chemical composition (dry matter, ash, crude fiber, crude protein, ether extract) was used as inputs and each individual aa was the output in each model. the results of this study showed that it is possible to predict aas with a simple analytical determination of proximate analysis. furthermore ann models could more effectively identify the relationship between aas and proximate analysis than linear regression model.