The Prediction of Somatic Cell Count Through Multilayer Perceptron of Deep Machine Learning

Abstract

The main objective of the research is to generate an alternative approach to classical techniques in the prediction of the somatic cell count (SCC), which is the gold standard indicator of subclinical mastitis. This approach involves using the physical properties of milk such as density, the temperature at fore milking (TFM), pH, and electrical conductivity (EC) with a feed-forward backpropagation multilayer perceptron (MLP) artificial neural networks (ANN) model, which is one of the widely used machine learning techniques. The performance of the model was assessed by test with cross-validation on data that was not introduced to the model before and compared to the classical linear model (multiple linear regression) as the control model. The findings showed that the model has satisfactory results in terms of loss and performance criteria (R2=0.95, RMSE=0.01; AIC=-338). The test model (ANN) had a higher performance (AIC=-338) than the control model (AIC=-240) created with the classical linear model despite using more parameters (81). Using big data from automated milking information—like estrus cycle, lactation stage, and milk yield—on supercomputers can improve the accuracy of performance assessments in dairy farming.

Keywords

• artificial neural networks (ANN)
• deep learning
• mastitis
• milk
• somatic cell count (SCC)