Evaluating machine learning algorithms based on thermal imaging and milk-based parameters to identify subclinical...

Area of Science:

• Veterinary Medicine
• Animal Science
• Machine Learning Applications

Background:

• Subclinical mastitis (SCM) is a prevalent udder infection in dairy cows, impacting milk production and quality.
• Early detection of SCM is crucial for effective herd management and economic viability.
• Current diagnostic methods can be labor-intensive and may not always provide timely results.

Purpose of the Study:

• To evaluate the efficacy of machine learning algorithms for identifying SCM in dairy cows.
• To assess the combined potential of thermal imaging and milk-based parameters for SCM detection.
• To compare the performance of different machine learning classifiers in this diagnostic context.

Main Methods:

• Utilized data from 194 Holstein dairy cows, collecting 776 quarter-level milk samples and thermal images.
• Analyzed milk for parameters and extracted skin temperatures from infrared images.
• Applied Random Forest Classifier (RFC), Logistic Regression, AdaBoost Classifier (ABC), and Naïve Bayes algorithms to identify SCM (defined as SCC > 200,000 cells/mL).

Main Results:

• RFC models integrating milk and thermal features achieved high accuracy (0.88), precision (0.88), and AUC (0.90), with lactose and SNF as key predictors.
• The AdaBoost Classifier using only thermal imaging data showed lower performance (accuracy 0.82, precision 0.67, AUC 0.56).
• Milk lactose and solids-not-fat (SNF) were identified as highly important features for SCM prediction.

Conclusions:

• Machine learning models integrating thermal imaging and milk-based parameters offer a promising approach for accurate SCM detection in dairy cows.
• Combining diverse data sources enhances diagnostic performance compared to using thermal imaging alone.
• Further research with larger datasets and refined imaging techniques is recommended to validate these findings and improve SCM management strategies.