AN APPLICATION OF MIXTURE EXPERIMENTS IN FORMULATION OF POULTRY FEED FOR MODELLING CHICK WEIGHT

Abstract

Poultry farming in Africa has significantly grown over the years. It is highly dependent on the economy and type of feed used. The quality, components, proportions, rations and storage of feeds is quite critical in poultry farming. The rising prices of the feed and the poor quality of the feed has become a concern to the poultry farmers due to low returns brought about by the low final weights among the birds. This study investigated how the theory of mixture experiments designs would be applied in the formulation of feed to come up with a model which farmers can apply in making their formulations to cut the cost of production and raise the profit margin. This study employed D-determinant, A-Average variance, E- Eigen value, and T-Trace optimality criterions in selecting an appropriate design between two major mixture experiments designs, simplex lattice and simplex centroid designs. The designs selection based on D-determinant, E-Eigen value, A-average variance and T-trace criterions rankings had equivalence at 1.5 but selection based on the fewer experimental runs led to the choice of simplex lattice mixture design. This design was applied in the formulation of poultry feed. The three components; whole maize meal, rice and wheat were being varied against the base or primary ingredient of proteins, vitamins, and minerals. The study varied the three major components, which are largely available and used by most poultry farmers in Kenya. Final weights data were analyzed using second order models. The effects of the feed proportions of the mixture on the final weights were analyzed using mixture regression method. The second order mixture models were used to create contour plots and response trace plot of the recorded weights and the used proportion of components. The contour plots proved useful in establishing desirable final weights and operating conditions. The optimum component proportions in the feed mixture for maximum weight gain were evaluated by response optimization analysis using the R software package as 1374.48g for whole maize component ,1367.01g for rice component and 1375.38g for wheat component .Given the three soya meal, cotton seed cake and fish meal protein bases, three second order linear models were obtained and compared in terms of F-statistics and coefficients of determination of 98.74%,99.24% and 99.98% respectively with adjusted R2 of 98.6%,99.16% and 99.98% respectively. The Fish meal base protein with the highest R2 of 99.98 and adjusted R2 of 99.98 was selected the most appropriate model to be recommended. The study concluded that fish meal protein gave the best model followed by Soya meal and cotton seedcake meal protein bases. Within the fish meal protein ,the varied proportions of carbohydrates, demonstrated that in the single component mixtures, component x1(whole maize) gave the highest final weight while binary blends, a combination of component x1(whole maize) and x3(wheat) gave the highest final weights among the other mixture combinations in the experiment.