KNN Algorithm Based on Locality Preserving

The distance metric plays an important role in K-nearest neighbor(KNN) algorithm. The traditional KNN algorithm usually employs the Euclidean distance. However, this distance treats all features equally and ignores the local intrinsic geometric structural characteristics of data. In this paper, following the basic idea of locality preserving projection(LPP), we firstly used the locality preserving within-class scatter matrix to propose a novel distance metric, then we developed a modified version of KNN called locality preserving K-nearest neighbor(LPKNN). The proposed method takes the local intrinsic geometric structural characteristics of data into full consideration. The experimental results indicate that the proposed algorithm can obtain higher classification accuracy in contrast with the KNN algorithm based on the Euclidean distance and the traditional Mahalanobis distance.