正则线性判别分析和最大散度判别分析的算法比较

我们给出了识别率偏差波动的计算公式，同时利用不同的参数在UCI的三个数据集上比较了正则线性判别分析和最大散度距离判别分析方法的识别性能。实验结果表明，在适当的参数下，正则线性判别分析的识别性能优于最大散度距离判别分析。另外，对于K近邻分类器中不同的K值，最大散度距离判别分析的识别率偏差波动要比正则线性判别分析的波动小。因此，在处理识别任务的实际应用中，对于一个稳定的识别方法，应该考虑识别率偏差波动。

A COMPARISON OF REGULARIZED LINEAR DISCRIMINANT ANALYSIS AND MAXIMUM SCATTER DIFFERENCE DISCRIMINANT ANALYSIS ALGORITHMS

A calculation formula of recognition rate deviation wave （RRDW） is introduced in this paper. Meanwhile, the recognition performance of regularized linear discriminant analysis （RLDA） and maximum scatter difference discriminant analysis 0VISD） methods were compared by using different parameters in three UCI data sets. The experimental results show that the recognition performance of RLDA is outperforms MSD under appropriate parameters. In addition, for different K values of K-nearest neighbor classifier （K-NNC）, the RRDW of MSD is smaller than RLDA. Therefore, in practical applications, RRDW should be considered as a stable method to handle recognition tasks.