基于聚类改进的Fisher与KNN判别分类算法对比研究

Comparative Study of Fisher and KNN Discriminant Classification Algorithms Based on Clustering Improvement

大米中矿物元素种类多(38种),为了快速、准确地筛选出判别大米产地的有效指标,在分析大米矿物元素含量相关性的基础上,进行了大米产地判别分类算法对比实验。首先通过R型聚类方法将大米矿物元素样本进行分块,然后从每一类中选取数据再进行Fisher与KNN判别分类对比,这样可以改进传统算法中剪辑样本带来的判别误差,又大大降低了无效的计算量。实验表明,基于聚类选出一种元素最优组合方案,采用23种矿物元素进行Fisher判别的分类率达86.76%,此方法准确高效地降低了计算机的运算量,提高了判别分类速度。

There are many kinds of mineral elements in rice(38 species),in order to quickly and accurately screen out the effective indicators of rice production,based on the analysis of the correlation of mineral content of rice,a comparative experiment of rice producing area discriminant classification algorithm was carried out.Firstly,the rice mineral element samples were segmented by the R-type clustering method,and then the data was selected from each of the clustering results and then compared with Fisher and KNN.This can improve the discriminant error caused by the clip samples in the traditional algorithm.It also greatly reduced the amount of invalid calculations.Experiments showed that based on clustering,an optimal combination scheme of elements was selected,and the classification rate of Fisher’s discriminant with 23 mineral elements was 86.76%.This method reduced the computational complexity of the computer accurately and efficiently,also improved the classification speed.