Prediction of the voluntary intake of grass silages by beef cattle 3. Precision of alternative prediction models
Prediction of the voluntary intake of grass silages by beef cattle 3. Precision of alternative prediction models
复制标题
DOI：
10.1017/s0003356100004931
复制
发表时间：
1990-06
影响因子：
--
通讯作者：
A. Rook;M. Dhanoa;M. Gill
中科院分区：
文献类型：
--
作者：
A. Rook;M. Dhanoa;M. Gill
使用均方预测误差对许多预测肉牛青贮饲料摄入量的新模型的精度进行了调查，并与两个先前发表的模型进行了比较（农业研究委员会，1980；Lewis，1981）。相对于之前的型号，新型号总体表现良好。新模型包括使用岭回归技术构建的一些模型，这些模型被证明比通过逐步最小二乘回归从相同估计数据获得的模型始终是更好的预测器。通过将最小二乘模型中的变量数量减少到使估计数据中的 R 2 最大化所需的变量以下，也可以获得更好的预测。具有最佳 R 2 的最小二乘模型的较差性能可能归因于估计数据中独立变量之间的共线性。大多数模型被认为相对于观察到的摄入量过度预测。这可能是由于测试数据和用于构建模型的估计数据之间的品种类型和动物管理方面的差异造成的，即模型与测试数据的使用涉及一定程度的外推。结论是，与基于最大化估计数据中的 R 2 的模型相比，岭回归和变量删除在摄入量预测方面向前迈出了积极的一步。然而，需要进一步的工作来阐明品种和饲养系统等因素对摄入量的影响，并阐明各种纤维测量在摄入量预测中的有用性。提出了许多利用一系列输入变量的新模型，从而允许在实际情况中灵活使用。
The precision of a number of new models for predicting silage intake by beef cattle was investigated with independent data using the mean-square prediction error and compared with two previously published models (Agricultural Research Council, 1980; Lewis, 1981). The new models generally performed well relative to the previous models. The new models included a number constructed using the technique of ridge regression which were shown to be consistently better predictors than the models obtained from the same estimation data by stepwise least-squares regression. Better prediction was also obtained by reducing the number of variables in the least-squares models below that required to maximize R 2 in the estimation data. The poor performance of the least-squares models with the best R 2 may be attributed to collinearity between the independent variates in the estimation data. Most of the models considered overpredicted relative to observed intakes. This may have been the result of differences in breed type and management of the animals between the test data and the estimation data used to construct the models, that is the use of the models with the test data involved a degree of extrapolation. It is concluded that ridge regression and deletion of variables offer a positive step forward in intake prediction compared with models based on maximizing R 2 in the estimation data. However, further work is needed to clarify the effect of factors such as breed and rearing system on intake and to clarify the usefulness of various fibre measures in intake prediction. A number of new models are proposed which utilize a range of input variables thus allowing flexibility in their use in practical situations.