Using Discriminant Analysis and Artificial Neural Network Models for Classification and Prediction of Fertility Status of Friesian Cattle

Abstract

Background & objectives: This study was undertaken to compare the accuracies of Discriminant analysis model (DA) and Artificial neural networks model (ANN) for classification and prediction of Friesian cattle fertility status by using its reproductive traits. Methods: Data was collected through field survey of 2843 animal records of Friesian breed belongs to El Dakhalia province farms, Egypt. Data was covering the period extended from 2010 to 2013. The samples of dairy production sectors were selected randomly. Data was collected from valid farm records or the structured questionnaires established by the researcher. Results: The results of classification accuracy indicated that the artificial neural network (ANN) model is more efficient than the discriminant analysis (DA) model in expressing overall classification accuracy and accuracies of correctly classified cases of fertility status for Friesian cattle. The results showed that The ANN models had shown the highest classification accuracy (93.6%) for year (2010) while, it was (79.9%) for DA. The comparison of overall classification accuracies clearly favored the supremacy of ANN over DA. The results also were confirmed by the areas under Receiver Operating Characteristic Curves (ROC) captured by ANN and DA. ROC curves are used mainly for comparing different discriminating rates. Areas under ROC curves were higher in case of ANN models across the different years compared to DA models. The differences in accuracies were also significant at 5% level of significance with p-value 0.005 by using Paired Sample t-test. From all of the above we can conclude that artificial neural network model was more accurate in prediction and classification of fertility status than a traditional statistical model (Discriminant analysis).