Abstract: The KNN classification algorithm is nonparametric, easy to understand and relatively efficient, and is widely used in many fields. In the traditional KNN algorithm, the Euclidean distance method considers the contribution of all the attributes of the sample as the same. But in fact, the contribution of different attributes of the sample is not necessarily the same. To solve this problem, an improved KNN algorithm based on analytic hierarchy process is proposed. In the improved algorithm, firstly, the weights of each attribute of the sample are calculated by using the analytic hierarchy process, and then the sample distance is calculated by using the weighted Euclidean distance, thereby classifying according to the weighted distance. In the experiment, with the increasing number of training samples, the efficiency of AHP-KNN algorithm is improved, and it is gradually better than the efficiency of the FCD-KNN algorithm and the traditional KNN algorithm. The simulation results show that the improved algorithm proposed can effectively improve the classification accuracy of the traditional KNN algorithm, and has certain theoretical and practical value.

Key words: KNN algorithm, Analytic hierarchy process, AHPKNN algorithm, FCD-KNN algorithm

摘要： KNN分类算法具有非参数性,易于理解且比较高效,被广泛应用于许多领域。传统的KNN 算法中的欧氏距离求法将样本所有属性的贡献视为相同,而实际上样本不同属性的贡献并不一定相同,为解决此问题,提出了一种基于层次分析法的改进KNN算法。在改进算法中,首先利用层次分析法计算样本各属性的权值,再采用加权的欧氏距离计算样本距离,根据样本的加权距离进行分类。实验中,随着训练样本的不断增加,AHP-KNN 算法的效率不断提高,并且逐步优于FCD-KNN算法和传统KNN 算法的效率。仿真结果表明,提出的改进算法有效提高了传统KNN算法的分类精确度,并具有一定的理论和实际应用价值。

关键词: KNN算法, 层次分析法, AHP-KNN算法, FCD-KNN算法