Milk yield prediction in Friesian cows using linear and flexible discriminant analysis under assumptions violations

Area of Science:

• Animal Science
• Data Science
• Agricultural Technology

Background:

• Novel technologies aid decision-making in livestock management.
• Friesian cattle milk production classification is crucial for optimal farm management.
• Predicting milk yield into low, medium, and high categories aids strategic planning.

Purpose of the Study:

• To evaluate linear discriminant analysis (LDA) and flexible discriminant analysis (FDA) for classifying Friesian cattle milk production.
• To compare the predictive performance of LDA and FDA models.
• To identify key predictors influencing milk yield categories.

Main Methods:

• Utilized 3793 lactation records from Friesian cows (2009-2020).
• Analyzed predictors: age at first calving, lactation order, days open, days in milk, dry period, calving season, 305-day milk yield, calving interval, and total breeding per conception.
• Compared LDA and FDA based on accuracy, sensitivity, precision, and F1-score.

Main Results:

• Flexible discriminant analysis (FDA) demonstrated superior performance over linear discriminant analysis (LDA).
• FDA achieved 82% classification accuracy, compared to LDA's 71%.
• FDA showed higher F1-scores (0.81-0.83) than LDA (0.667-0.79).

Conclusions:

• Flexible discriminant analysis (FDA) is more robust than linear discriminant analysis (LDA) when assumptions are violated.
• Both LDA and FDA are effective tools for interpreting and predicting livestock datasets.
• FDA offers improved accuracy for classifying cattle milk production.