This study is aimed at describing the lactation curve using an empirical (Wood and Wilmink), stochastic (Djisktra and MilkBot) and artificial intelligent (Neural network) models to developing an equations that drive milk production pattern of different genotype of cows under a low input production system of Nigeria. Five functions were fitted to test day milk yield records of Holstein, HolsteinXBunaji, Jersey and HolsteinXBunaji cows. The fit of the functions was evaluated using adjusted coefficient of determination (R2) and their predictive abilities were compared using root mean square error and bayesian information criterion. The empirical function of Wilmink had convergence failure with atypical lactation curve. Stochastic function of Djisktra and MilkBot models, empirical function of Wood and artificial neural network were all good predictors of lactation curve (Adj R2=>80%) with the exception of Wilmink (Adj R2=62%) in Holstein and Djisktra (Adj R2=62%) in HolsteinXBunaji. In Jersey and JerseyXBunaji cows, all the models were efficient (Adj R2=>80%; 75%) in reconstructing the extended phase of lactation curve. Artificial neural network consistently showed superior fit across the genotypes except for Holstein where MilkBot was the best. The NN function demonstrated higher adaptability to various data characteristics than all the models and could be used in situations where animal recording is not consistently practised and where recording of animal performance is routinely practised. Milkbot function had high data requirements in Holstein cows, which restricts it to dairy systems with consistent recording, despite easy physiological interpretation of its parameters.

Keywords: Keywords: Emperical, stochastic, artificial neural network, genetics, environment