Prediction of immunoglobulin g in lambs with artificial intelligence methods

Abstract

The health, mortality and morbidity rates of neonatal ruminants depend on colostrum quality and the amount of Immunoglobulin G (IgG) absorbed. Computer-aided estimates are important as measuring IgG concentration with conventional methods is costly. In this study, artificial neural network (ANN), multivariate adaptive regression splines (MARS), support vector regression (SVR) and fuzzy neural network (FNN) models were used to predict the serum IgG concentration from gamma-glutamyl transferase (GGT) enzyme activity, total protein (TP) concentration and albumin (ALB). The correlation between parameters was examined. IgG positively correlated with GGT and TP and negatively correlated with ALB (R = 0.75, P575 mg/dL (P=0.02), GGT >191 mg/dL (P55 g/L (P<0.001) were determined for healthy. It has been observed that the FNN is the most successful method for the prediction of IgG value with a correlation coefficient (R) of 0.98, root mean square error (RMSE) of 234.4, and mean absolute error (MAE) of 175.8.

Keywords

• artificial neural network
• decision tree
• fuzzy neural network
• immunoglobulin g
• multivariate adaptive regression splines
• support vector regression