Machine learning classification of breeding protocol descriptions from Canadian Holsteins

Area of Science:

• Animal Science
• Genetics
• Data Science

Background:

• Dairy farmers use timed artificial insemination (TAI) to optimize cow pregnancies.
• TAI can obscure natural fertility, potentially biasing genetic evaluations for fertility traits.
• Current breeding protocol records lack uniformity and specificity.

Purpose of the Study:

• To investigate the efficacy of supervised machine learning algorithms in classifying dairy cattle breeding protocols.
• To differentiate between TAI and non-TAI breeding methods using farmer-recorded descriptions.
• To enable more accurate genetic evaluations for fertility by accounting for breeding management.

Main Methods:

• Utilized 8 supervised machine learning algorithms.
• Classified 1,835 unique breeding protocol descriptions from 981 dairy herds.
• Evaluated algorithm performance using metrics like Matthews correlation coefficient and F1-score.

Main Results:

• Stacking classifier algorithms achieved the highest performance (MCC = 0.94 ± 0.04, F1-score = 0.96 ± 0.03).
• Several machine learning models demonstrated robust performance in identifying TAI protocols.
• The study confirmed the feasibility of using machine learning for breeding protocol classification.

Conclusions:

• Machine learning algorithms can reliably distinguish TAI from other breeding protocols in dairy cattle.
• This classification facilitates unbiased genetic evaluations of natural fertility.
• Improved data management through machine learning supports better breeding decisions in dairy farming.