基于多分割和口袋方法的最小二乘分类算法

Least squares classification algorithm based on multi-segmentation and pocket

提出了一种改进的最小二乘分类算法,该算法首先利用最小二乘算法对两类数据分类,然后计算每类的中心点,过中心点作已得到的分类线(面)的平行线(面),保留所作平行线(面)之间及线(面)上的数据,剔除其余数据,对剩余数据重新利用最小二乘算法分类,如此循环。在循环过程中利用口袋方法记录下具有最好的分类效果的分类线(面),循环结束后口袋中即为最佳分类线(面)。实验结果表明,该算法有效的解决了原有最小二乘分类算法的缺陷,有着良好的分类效果。

An improved least squares algorithm is presented.The two kinds of data are classified by least squares algorithm first,and then the mean points of the two kinds of data are calculated.After drawing the parallel lines（planes） of the classification line（plane） through the mean points respectively,some data which between the lines（planes） or on the lines（planes） are preserved,the others are deleted.The rest data are re-classified by the least squares algorithm.In these iterative loops the classification line（plane） which has the best classification result is stored in a pocket.At last the classification line（plane） is founded.Experimental results prove the avoidance of the original algorithm shortcoming and the higher performance of the new least squares algorithm.