一种基于改进分布估计算法的统计模型选择

A Statistic Model Selection Based on the Improved Estimation of Distribution Algorithm

回归模型是数据处理中的经典分析方法,在许多领域都有重要应用。回归模型的模型选择的常用方法是专家选择和完全模型法,但两者都过于依赖主观经验,影响系统的运行效果。针对选择回归模型问题的特点,在分布估计算法基础上设计了相应的概率矩阵,以赤池信息准则(AIC)和贝叶斯信息准则(BIC)作为模型选择的度量标准,提出了解决统计模型选择问题的改进分布估计算法。该算法基于遗传算法变异思想构造变异算子,基于免疫机理设计选择策略。最后,通过实例,将新算法与传统的遗传算法,以及经典的模型选择方法进行仿真比较,得出新算法在AIC、BIC值等各种指标上效果提高。

Regression model is a classical analysis method in data processing.It has numerous applications in many fields.The most commonly-used methods are expert selection and complete model method.However,they are too dependent on the subjective experience.According to the characteristics of regression model selection problem,we design the corresponding probability matrix.Then,by combining Akaike Information Criterion（AIC）and Bayesian Information Criterion（BIC）,we present an improved estimation of distribution algorithm for solving statistical model selection problem.It constructs mutation operator based on the idea of mutation of GA.Moreover,the selection policy is designed on the basis of the immune mechanism.Finally,we use computer simulation to compare this improved EDA,traditional genetic algorithm and classical algorithm for model selection and to show that the new EDA has superiority in AIC,BIC and other various indices.