| hrough the use of an artificial neural network the analysis of reproductive performance in high producing dairy cattle using startup lactation information was evaluated. Data was obtained from monthly DHIA tapes and used to build a database utilizing current and historical lactation records from both multiparous and primiparous cows.;A preliminary statistical analysis was conducted to identify those variables that showed the largest influence upon days open. These variables were then used to define the training datasets. Training was performed with two different neural network structures, one hidden layer versus two hidden layers. Initial neural network results showed that neither of the neural networks improved training and testing results (from the start of training).;A statistical analysis was performed to identify those parameters that have the largest influence upon days open. These parameters (previous days open, previous lactation last milk weight, first milk weight, change in milk production from test day 1 to test day 2, change in milk production from test 2 and test day 3, days dry, difference between milk fat and protein for 1st test day, average milk weight for test day 2 and test day 3, average milk weight for test day 1 and test day 3, and protein average for test day 1 and test day 3) were used to define new training datasets. Both neural networks performed at a higher level with the number of correctly estimated days open increasing dramatically (92% correct). However, the margin in which days open is estimated allowed a plus or minus 30-day range from the actual days open. When the days open margin was reduced to a smaller range of days, the effect was poor training and testing results. These results were attributed to numerous contradictions in the data. An analysis of variance was performed for primiparous and multiparous cows with the goal of identifying those variables that statistically differed across days open intervals. Only previous lactation days open, days dry and protein percent showed statistical significance (P... |
