一种基于马氏距离的线性判别分析分类算法

A Linear Discriminant Analysis Classification Algorithm Based on Mahalanobis Distance

对于一个特定的模式识别问题,表达和识别模式的特征具有不同的形式,它们在物理意义上是完全不同的,而且在数量级具有很大差别。该文提出了一种基于马氏距离的线性判别分析分类算法,选取判别函数为马氏距离,可以适用于具有不同类型特征值的分类问题。将该算法应用于UC I中C red it-A、C red it-G、Iris和Veh ic le四个数据库的分类,并采用K次交叉验证方法进行实验。从实验结果中可知,与ENTROPY算法和C4.5(8)算法分类效果相比较,该文所提出的线性判别分析算法计算简单,识别率较高,是一种实际可行的分类算法。

The features for a pattern recognition problem have different forms, physical meanings and quantity levels. A linear discriminant analysis classification algorithm based on Mahalanobis distance is proposed in this paper. In the algorithm, the Mahalanobis distance is chosen as discriminant function. The algorithm can be applied to classification problem having several kinds of property. The Credit - A, Credit - G, Iris and Vehicle databases from UCI repository are tested and K - cross validation is adopted. The experiment results show that compared with the ENTROPY and C4.5（8） algorithm, the computation amount of the linear discriminant analysis algorithm is less, and its ration of recognition is higher. It is a feasible algorithm.