An improved feature-weighted method based on K-NN

In this paper, an improved feature-weighted method is proposed to improve the classification performance of K-NN. In cluster analysis, certain features of samples may exhibit higher relevance than others. To address this problem, feature weighting is one of the effective methods. On the basis of guaranteeing the minimum distance between different classes of samples, a feature-weighted method is established to minimize the maximum distance between the similar samples, there are three advantages: 1. it can be expressed as a simple linear programming problem; 2. does not depend on the scale of different features; 3. automatically reduce the data dimension in the process of feature weighting. However, this method has a distinct defect, that is, the robustness of the algorithm is not strong, and it is very sensitive to the noise sample. In order to solve this problem, this paper uses the local density to weight the feature based on the distribution characteristics of the samples around a certain sample, and establishes the feature-weighted method based on the weighted maximum distance minimization, which can effectively reduce the influence of the noise sample. The real data sets of UCI database are used to verify the proposed method. The experimental results show that compared with the prior method, the algorithm's robustness and the ability to anti-noise has been effectively improved.