一个实际回归模型的优化选择

Optimal Choice of a Practical Regression Model

本文是对一个实际问题作最优回归模型的选择。对本文所论及的问题 (见 2应用举例 ) ,作者所见到的均以指数回归模型或双曲回归模型作为最优的模型 ,可通过回归拟合、显著性检验与区间预测及方差分析发现更优的回归模型应是反比例回归模型。而且 ,作者还发现双曲线和指数回归模型的结果不能严格地满足平方和分解定理 ,反比例回归模型则能精确满足。

Optimal choice of a practical regression model is discussed in this paper. As far as the practical problem in this paper is concerned, all consider exponential or hyperbola regression model as the optimal one. In fact, the author finds that inverse proportion curve regression model should be the optimum by regression fit, significance test, interval prediction,and variance analysis. Both exponential and hyperbola curve regression model can't satisfy the quadratic sum decomposition theorem strictly, but the inverse proportion regression model can satisfy it well.