Optimized feed formulation remains a significant challenge in smallholder dairy systems in sub-Saharan Africa due to seasonal variability in feed resources and limited access to nutrition expertise. An automated feed formulation algorithm was previously developed to address these gaps using linear programming and nutrient requirements for dairy cattle. This study aimed to validate the algorithm’s predictive accuracy and precision by comparing model-predicted milk yields with observed yields from 15 lactating cows, comprising six Holstein Friesians (HF) and nine Holstein Friesian–Zebu crosses (HFZ), managed under smallholder conditions in Malawi. Cows were individually fed tailored diets generated by the algorithm over a 30-day period in a challenge-feeding approach. Observed milk yield, body weight, body condition score, and milk composition were recorded. Predictive performance was assessed using the Concordance Correlation Coefficient (CCC), Root Mean Square Error (RMSE), Mean Bias (MB), and coefficient of determination (R²). Results showed moderate CCC (0.772) and R² (0.899) when predictions were aggregated for the entire trial period, while daily RMSE and MB values indicated a mean overestimation of 2.96 kg. Observed milk yield was also significantly influenced by days in milk and feed intake, suggesting that both farm-level management and individual cow variation contributed to differences between predicted and observed production. The tool demonstrated reasonable predictive ability for milk yield and potential practical field application; however, the results highlight areas requiring further refinement to improve model precision.
Mtawali et al. (2026) studied this question.