Comparison of Nonlinear Models and Artificial Neural Networks for Modelling Lactation Curves in Holstein-Friesian × Bunaji Crossbred Dairy Cows
Makchit Galadima
Department of Theriogenology and Production, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
David Oludare Omoniwa
Department of Veterinary Medicine, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
Collins Idachaba Udagbene
Department of Animal Production, Faculty of Agriculture, University of Jos, Plateau State, Nigeria.
Kassam Isuwa Rotji
Department of Theriogenology and Production, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
Mary Foluke Oladipo
Department of Theriogenology and Production, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
Anas Usman Musa
Malcomines Dairy Farm, Babale, Jos North LGA, Plateau State, Nigeria.
Imaben Opaluwa Grace-Kuzayed
Department of Theriogenology and Production, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
Enupe Gideon Ogba
Department of Theriogenology and Production, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
Oludayo Micheal Akinsola *
Department of Theriogenology and Production, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
Babalola Samuel Adeola
Department of Veterinary Surgery and Radiology, Faculty of Veterinary Medicine, University of Jos, Plateau State, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
The dairy industry in Nigeria faces chronic low productivity, with indigenous Bunaji cattle yielding only 0.5–3.0 L/day and national output meeting just 35–40% of demand. Crossbreeding with Holstein-Friesian has shown heterosis, particularly in the Jos Plateau’s cool highland climate, yet the absence of individual-level test-day records limits genetic evaluation and herd management. This study compared the predictive performance of three parametric non-linear models (Wood, Wilmink, and Ali and Schaeffer) and an artificial neural network (ANN) in modelling lactation curves of Holstein-Friesian × Bunaji crossbred dairy cows under intensive management. The dataset comprised 21 complete lactations of Holstein-Friesian × Bunaji crossbred cows, each with 35 weekly milk yield observations, resulting in 735 potential test-day records used for lactation curve modelling. Derived parameters included peak yield, time to peak, persistency index, and average yield. Model efficiency was assessed using adjusted R2, root mean square error (RMSE), Akaike information criterion (AIC), and Bayesian information criterion (BIC). The Ali and Schaeffer model provided the superior statistical fit (Adj. R2 = 0.97, RMSE = 0.25, AIC = −87.5, BIC = −79.7), followed by Wood’s model. The ANN yielded the highest biological realism (peak yield = 8.40 L at week 4; persistency index = 0.54) but the weakest goodness-of-fit. All models produced similar average daily yields (≈6.49 L), with peaks ranging 7.72–8.40 L and persistency indices 0.47–0.54. These findings demonstrate that classical parametric models such as Ali and Schaeffer remain the most parsimonious and accurate for reconstructing lactation curves from limited intensive data in Nigeria. ANN offers complementary value for early prediction of key parameters. Accurate curve modelling will support context-specific breeding programmes, genetic evaluation, and sustainable dairy intensification on the Jos Plateau. Precise lactation curve modeling enhances estimation of key parameters (peak yield, persistency, and total lactation milk production), supporting improved herd management, genetic evaluation, and breeding strategies to boost productivity in crossbred cattle under similar production systems.
Keywords: Lactation curve modelling, Holstein-Friesian x Bunaji crossbred dairy cows, artificial neural networks, milk yield persistency, non-linear regression