Comparison of linear discriminant analysis, support vector machine and artificial neural network in classifying Nigerian local turkeys based on plumage colours using biometric traits

The ability of linear discriminant analysis (LDA), support vector machine (SVM), and artificial neural network (ANN) models to differentiate biometric traits of Nigerian local turkeys was investigated in this study. The biometric traits (bodyweight, body length, breast girth, thigh length, shank length, keel length, wing length, and wingspan) in 200 (20-week-old) turkeys were measured. Seventy percent of the datasets were used to train the three models, with the remaining 30% being used to test their performance. All biometric traits were positively associated, with strong correlation values for several pairs of traits. In the testing dataset (Lavender = 30.0%, Black = 51.9% and White = 65.5%), the LDA had lower classification efficiency than in the training dataset (Lavender = 55.2%, Black = 43.4%, and White = 65.5%), indicating that the training model was not efficient in classification at the testing stage. In comparison to the training dataset (Lavender = 100.0%, Black = 87.3% and White = 98.2%), the SVM showed low classification efficiency for the testing dataset (Lavender = 70.0%, Black = 76.0% and White = 64.0%). However, in ANN, there was no variation in classification efficiency between the testing and training datasets (Lavender = 100.0%, Black = 100.0% and White =100.0%). In categorizing turkey plumage colours, the ANN model is the most powerful, followed by SVM. When the dataset's normality or multi-colinearity is broken, we propose using an ANN model rather than a standard model like the LDA for classification of biometric traits of Nigerian local turkeys.