Using multiple regression, Bayesian networks and artificial neural networks for prediction of total egg production in European quails based on earlier expressed phenotypes

The prediction of total egg production (TEP) potential in poultry is an important task to aid optimized management decisions in commercial enterprises. The objective of the present study was to compare different modeling approaches for prediction of TEP in meat type quails (Coturnix coturnix coturnix) using phenotypes such as weight, weight gain, egg production and egg quality measurements. Phenotypic data on 30 traits from two lines (L1, n=180; and L2, n=205) of quail were modeled to predict TEP. Prediction models included multiple linear regression and artificial neural network (ANN). Moreover, Bayesian network (BN) and a stepwise approach were used as variable selection methods. BN results showed that TEP is independent from other earlier expressed traits when conditioned on egg production from 35 to 80 days of age (EP1). In addition, the prediction accuracy was much lower when EP1 was not included in the model. The best predictive model was ANN, after feature selection, showing prediction correlations of r=0.792 and r=0.714 for L1 and L2, respectively. In conclusion, machine learning methods may be useful, but reasonable prediction accuracies are obtained only when partial egg production measurements are included in the model.

Bootstrapped neural network models for analyzing the responses of broiler chicks to dietary protein and branched chain amino acids

Faridi, A., Golian, A., Heravi Mousavi, A. and France, J. 2014. Bootstrapped neural network models for analyzing the responses of broiler chicks to dietary protein and branched chain amino acids. Can. J. Anim. Sci. 94: 79–85. Reliable prediction of avian responses to dietary nutrients is essential for planning, management, and optimization activities in poultry nutrition. In this study, two bootstrapped neural network (BNN) models, each containing 100 separated neural networks (SNN), were developed for predicting average daily gain (ADG) and feed efficiency (FE) of broiler chicks in response to intake of protein and branched chain amino acids (BCAA) in the starter period. Using a re-sampling method, 100 different batches of data were generated for both the ADG and FE sets. Starting with 270 data lines extracted from eight studies in the literature, SNN models were trained, tested, and validated with 136, 67, and 67 data lines, respectively. All 200 SNN models developed, along with their respective BNN ones, were subjected to optimization (to find the optimum dietary protein and BCAA levels that maximize ADG and FE). Statistical analysis indicated that based on R 2, the BNN models were more accurate in 76 and 56 cases (out of 100) compared with the SNN models developed for ADG and FE, respectively. Optimization of the BNN models showed protein, isoleucine, leucine, and valine requirements for maximum ADG were 231.80, 9.05, 14.03 and 10.90 g kg−1 of diet, respectively. Also, maximum FE was obtained when the diet contained 232.30, 9.07, 14.50, and 11.04 g kg−1 of protein, isoleucine, leucine, and valine, respectively. The results of this study suggest that in meta-analytic modelling, bootstrap re-sampling algorithms should be used to better analyze available data and thereby take full advantage of them. This issue is of importance in the animal sciences as producing reliable data is both expensive and time-consuming.

Study of broiler chicken responses to dietary protein and lysine using neural network and response surface models

1. In this study, neural network (NN) and response surface (RS) models were developed to investigate the response [average daily gain (ADG) and feed efficiency (FE)] of young broiler chickens to dietary protein and lysine. For this purpose, data on their responses to dietary protein and lysine were extracted from the literature and separate NN and RS models were constructed. 2. Comparison between the NN and RS models revealed higher accuracy of prediction with the NN models compared to the RS models. In terms of R (2) values, the NN models developed for both ADG (R (2) = 0.923) and FE (R (2) = 0.904) were far superior to the RS models (R (2) for ADG = 0.511; R (2) for FE = 0.67). This suggests that the NN models can serve as an alternative option to conventional regression approaches including use of RS models. 3. Optimisation of the NN models developed for response to protein and lysine showed that diets containing 220.7 (g/kg of diet) protein and 12.85 (g/kg of diet) lysine maximise ADG, whereas maximum FE is achieved with diets containing 241.3 and 13.12 (g/kg) protein and lysine, respectively. Based on the optimisation results, optimal dietary protein and lysine concentrations for maximum FE in broiler chickens during the starting period are higher than for ADG.

A comparison of neural network models, fuzzy logic, and multiple linear regression for prediction of hatchability

Application of appropriate models to approximate the performance function warrants more precise prediction and helps to make the best decisions in the poultry industry. This study reevaluated the factors affecting hatchability in laying hens from 29 to 56 wk of age. Twenty-eight data lines representing 4 inputs consisting of egg weight, eggshell thickness, egg sphericity, and yolk/albumin ratio and 1 output, hatchability, were obtained from the literature and used to train an artificial neural network (ANN). The prediction ability of ANN was compared with that of fuzzy logic to evaluate the fitness of these 2 methods. The models were compared using R(2), mean absolute deviation (MAD), mean squared error (MSE), mean absolute percentage error (MAPE), and bias. The developed model was used to assess the relative importance of each variable on the hatchability by calculating the variable sensitivity ratio. The statistical evaluations showed that the ANN-based model predicted hatchability more accurately than fuzzy logic. The ANN-based model had a higher determination of coefficient (R(2) = 0.99) and lower residual distribution (MAD = 0.005; MSE = 0.00004; MAPE = 0.732; bias = 0.0012) than fuzzy logic (R(2) = 0.87; MAD = 0.014; MSE = 0.0004; MAPE = 2.095; bias = 0.0046). The sensitivity analysis revealed that the most important variable in the ANN-based model of hatchability was egg weight (variable sensitivity ratio, VSR = 283.11), followed by yolk/albumin ratio (VSR = 113.16), eggshell thickness (VSR = 16.23), and egg sphericity (VSR = 3.63). The results of this research showed that the universal approximation capability of ANN made it a powerful tool to approximate complex functions such as hatchability in the incubation process.

Evaluating the egg production of broiler breeder hens in response to dietary nutrient intake from 31 to 60 weeks of age through neural network models

Faridi, A., Golian, A. and France, J. 2012. Evaluating the egg production of broiler breeder hens in response to dietary nutrient intake from 31 to 60 weeks of age through neural network models. Can. J. Anim. Sci. 92: 473–481. The aim of this study was to evaluate the response of broiler breeder hens in terms of egg production to dietary nutrient intake. Using neural network (NN) models and breaking down the collected data from 98 commercial broiler breeder houses into 3-wk intervals, 10 NN-based models were developed from 31 to 60 wk of age. The data lines were divided into two random subsets of training (n=64) and testing (n=34) sets. The variables of interest for developing the models were metabolizable energy (ME; kcal bird−1 d−1), and crude protein (CP), total sulphur amino acids (TSAA), lysine (Lys), calcium (Ca) and available phosphorus (AP), all in g bird−1 d−1. The random optimization algorithm was applied to the constructed models to find the optimal level of the input variables which maximized egg production during the different intervals. The high R 2 values in all the developed models for both the training and testing sets indicate the accuracy of NN-based models in estimating egg production. The optimization results revealed that breeder hens consuming 485, 473, 471, 466, 460, 452, 448, 442, 437 and 445 kcal of ME bird−1 d−1 showed the highest egg production during the 10 consecutive 3-wk intervals from 31 to 60 wk of age, respectively. Moreover, the optimal performance of hens required the following average intakes from 31 to 60 wk of age (g bird−1 d−1): CP: 23.7; TSAA: 1.05; Lys: 1.07; Ca: 4.91; and AP: 0.58. The results show that energy (kcal bird−1 d−1) and other nutrient requirements (g bird−1 d−1) of broiler breeder hens from 31 to 60 wk of age do not change in consort together with age; therefore using different diets with different dietary nutrient levels during the production cycle may help the nutritionists better meet the requirements of broiler breeder hens. Based on the present study, it appears that company guideline recommendations may underestimate the nutrient requirements of hens during these weeks when egg production is declining gradually.