Application Prospect of Artificial Neural Network in Dairy Cow Mastitis Prediction

Abstract

This research aimed to study the prospect of application of artificial neural network (ANN) on a prediction of mastitis in dairy cows by using factors from traditional and alternative methods. Factors of traditional method used in this study were pH and electrical conductivity (EC.) of raw milk and factor of alternative method was thermal imaging of cow udder. Samples were collected from 5 farms in Nakhon Pathom province. From 645 cows, 112 cows were specifically collected for thermal image data and raw milk for determination of pH and EC. These 3 factors were used as input data for the ANN model. The hidden layer consists of 3 layers and the first, second and third layer consists of 4, 3 and 2 nodes, respectively. The output has 1 answer which was classified into three groups including non-mastitis, sub-clinical mastitis and clinical mastitis. In this research, the dataset was divided into training data (70%), testing data (30%), and validation data (100%) using somatic cell count and California Mastitis Test (CMT). The evaluation of prediction accuracy using the artificial neural network model showed an accuracy in learning was 85.90%, the testing has an accuracy of 79.41% and the validation of prediction has an accuracy of 85.71% indicating that ANN model using factors from traditional method (pH and EC) and factor from alternative method with thermal image is an efficient tool for mastitis prediction and potential to develop the system for evaluation of milk Quality determination using rapid method.