Enhanced mastitis severity classification in dairy cows using DNN and RF: A study on PCA and correlation-based feature selection

Abstract

Mastitis presents a critical challenge in dairy farming, significantly impacting both animal health and economic stability. This study advances mastitis severity classification in Holstein-Friesian cows through an interdisciplinary approach that integrates advanced machine learning methods within a veterinary context, shifting focus from traditional disease detection or diagnosis to refined severity classification, which is critical for aligning treatment strategies with disease progression and optimizing intervention timing. Clinical measurements from 1,886 Holstein-Friesian dairy cows, aged 3-4 years, across 21 parameters were analyzed. Both Deep Neural Networks (DNN) and Random Forests (RF) were utilized, with dimensionality reduction applied through Principal Component Analysis (PCA) and correlation-based feature selection to retain essential features and enhance computational efficiency. Key parameters in the RF and DNN models were carefully selected to ensure robust performance across diverse datasets. By categorizing mastitis into acute, acute with bloody milk, and peracute stages, this study addresses unique diagnostic challenges within veterinary medicine. The RF and DNN models achieved superior accuracy, ranging from 97% to over 98%, surpassing traditional diagnostic methods in precision, recall, and F1 metrics. These findings highlight the broader applicability of machine learning advancements beyond veterinary sciences, demonstrating the potential for interdisciplinary research across veterinary medicine, computer science, and engineering to drive innovation in automated diagnostics and resource management in dairy farming. The research was conducted using the MATLAB® Deep Learning App.