A New Effective Kernel-Based k-Nearest Neighbor Classification Method

The Z-KNN model is upgraded to the proposed model based on a Gaussian kernel. In essence, the Gaussian kernel provides a way to gauge how similar samples are, which allows for better clustering of similar samples in a space that portrays “similarity” and is thus linearly separable. Based on the Euclidean distance, the Z distance can be seen as an enhancement of the Euclidean distance. In contrast to Euclidean distance, Z distance makes the distance between data belonging to distinct classes larger than between those belonging to the same class. The input data are transferred to a high-dimensional feature space by a Gaussian kernel method, and nonlinear mapping and Z-KNN are performed in this space. Experiments demonstrate that the proposed model produces superior classification results on nonlinear data sets.