关键词: 多维数据, 判别分析, 非参数统计, 核函数, 概率密度

Abstract: Discriminant analysis is widely used in data mining and recognition. How to make full use of the information of training sets, and how to improve the algorithm of discriminant rules and reduce the rate of misjudgement has always been the focus for many researches. In some traditional algorithms, the distribution type of data is often assumed firstly, but the structures of multidimensional data often violate the assumptions and lead to a higher rate of misjudgment. Aiming at such problems, this paper proposes to establish discriminant rules by the algorithm of nonparametric kernel density, and carries out empirical analysis through Iris and Seeds data. The results show that compared with the existing discriminant analysis algorithms, the proposed algorithm uses the information of data more fully, and significantly improves the accuracy of the multidimensional data. At the same time, this algorithm is not restricted by the distribution assumption, so it has wide applicability.

Key words: multidimensional data, discriminant analysis, nonparametric statistics, kernel function, probability density