A Kernel-Based Local Mean k-Nearest Centroid Neighbor Method for Classification

Abstract

Listen

Translate article

K-nearest neighbor criterion has a broad research and application background in pattern recognition, machine learning, data mining and other fields. In recent years, a series of extended methods have been proposed, such as k-nearest centroid neighbor(KNCN), local mean-based k-nearest centroid neighbor (LMKNCN), etc. This is because more scholars have studied k-nearest neighbor (KNN) method. Some scholars have proposed kernel-k-nearest neighbor(kernel-KNN) method by combining the support vector machine (SVM) and KNN method. Compared with the traditional KNN method, these methods have a better classification effect. However, the classification effect of LMKNCN method is inferior to that of regular sample distribution in the case of the random sample distribution. And kernel method can effectively solve indivisible linear problems by transforming the problem into high-dimensional space. In this paper, the kernel method is combined with LMKNCN. A new method, kernel-local mean-based k-nearest centroid neighbor (kernel-LMKNCN), is obtained by nucleating the original LMKNCN methods. Experimental results show that the proposed kernel-LMKNCN methods are superior to traditional KNCN, LMKNCN and kernel-KNN methods in classification accuracy.