Background The application of novel technologies is now widely used to assist in making optimal decisions. This study aimed to evaluate the performance of linear discriminant analysis (LDA) and flexible discriminant analysis (FDA) in classifying and predicting Friesian cattle’s milk production into low ( 4500 kg), medium (4500–7500 kg), and high ( 7500 kg) categories. A total of 3793 lactation records from cows calved between 2009 and 2020 were collected to examine some predictors such as age at first calving (AFC), lactation order (LO), days open (DO), days in milk (DIM), dry period (DP), calving season (CFS), 305-day milk yield (305-MY), calving interval (CI), and total breeding per conception (TBRD). Results The comparison between LDA and FDA models was based on the significance of coefficients, total accuracy, sensitivity, precision, and F1-score. The LDA results revealed that DIM and 305-MY were the significant (P < 0.001) contributors for data classification, while the FDA was a lactation order. Classification accuracy results showed that the FDA model performed better than the LDA model in expressing accuracies of correctly classified cases as well as overall classification accuracy of milk yield. The FDA model outperformed LDA in both accuracy and F1-score. It achieved an accuracy of 82% compared to LDA’s 71%. Similarly, the F1-score improved from a range of 0.667 to 0.79 for LDA to a higher range of 0.81 to 0.83 for FDA. Conclusion The findings of this study demonstrated that FDA was more resistant than LDA in case of assumption violations. Furthermore, the current study showed the feasibility and efficacy of LDA and FDA in interpreting and predicting livestock datasets.