DEVELOPMENT OF FEATHER WEIGHT PREDICTION MODEL USING ARTIFICIAL NEURAL NETWORK OF BROILER CHICKENS
Despite low cost of chicken feathers, abundant availability, worldwide applicability and unique properties, there hasn’t been any economic means of utilizing it in Nigeria. Based on this background, this research attempted to develop a model using neural network for prediction of chicken feather weights in Nigeria. The model used body weights at different stages in predicting the feather weights. An Experiment was conducted by raising 1000 chicken from day old to 7 weeks. The Chicken’s body weights were measured 7 times at an interval of 1 week. These weights were used as input data for developing the model. A two layer feed forward network with 4 sigmoid hidden Neurons and 1 Output Neuron were used. The data was divided into 3 sets (70% training sets, 15% testing sets and 15% validating sets). The network was build using MATLAB Software and was subjected to Levenberg-Marquadrt algorithm. The model Simulink equation was generated using the MAPMINMAX preprocess link and the output was calculated from the MAPMINMAX reverse mask. A good agreement between the predicted weights and the experimental weights was achieved with 0.8546 as the coefficient of determination (r2) and the sensitivity results ranged from 0.04 to 0.6 for the validated model. In model validation, results revealed (r2) value and MSE of 0.88033 and 0.0007679 respectively. Therefore the feather weights of broiler chicken in metropolitan Kano can be best predicted using the developed ANN model. Modeling the production of chicken feather would enhance the economic means of its utilization.
Authors : Nashe, A. J., Dangora, N. D., Bature, A. and Suleiman, J.
Category : Open Access Volume (Issue) : 7(1) Date Uploaded : 4th July 2021