基于Chernoff上界的多类问题特征选择优化迭代算法

本文提出正态分布条件下面向不同分布的多类问题基于Chernoff上界的特征选择优化迭代算法.该算法由两重迭代组成首先在设定的原始空间和特征空间Chernoff参数s条件下,通过解矩阵方程的迭代算法求得变换矩阵的最优解;然后,在变换矩阵确定的特征空间中搜索最佳的参数s使错误概率上界最小;最后采用折半法修正设定的Chernoff参数s及其迭代步长.通过分析和实例可见基于Chernoff上界特征选择是面向不同分布的多类问题的最佳特征选择方法.

AN OPTIMAL RECURSIVE ALGORITHM FOR Chernoff BOUND FEATURE SELECTION IN MULTICLASS PROBLEMMS

An optimal recursive algorithm for Chernoff bound feature selection of multiclass problem and normal multidistribution oriented is presented. This algorithm is twofold. Setting the parameter 's' in the feature and original space, the optimal solution of transformation matrix can be found. The optimal set of parameter's' can be searched by minimizing the upper bound of error probability. Then, the setting parameter's'and recursive step should be corrected to close the difference between the setting and optimal 's'. The theoretical analysis and experimental results show that the performance of proposed algorithm is superior to the performance of any previous one.