K-nearest neighbor rule(KNN) is a simple and powerful classification algorithm. In this article, we develop a representation-based pseudo nearest neighbor rule(RPNN), which is based on the idea of pseudo nearest neighbor rule(PNN). In the proposed RPNN, a query point is represented as a linear combina‐tion of all the training samples in each class, and we use k largest representation coefficients to determine k nearest neighbors per class instead of Euclidean distance. Then, we design the pseudo nearest neighbor of the query point per class and compute the distance between the query point and the pseudo nearest neighbor. The query point belongs to the class which has the closest representa‐tion-based pseudo nearest neighbor among all classes. Experimental results on twelve data sets have demonstrated that the proposed RPNN classifier can improve the classification accuracy in the small sample size cases, compared to the traditional KNN-based methods.
K-nearest neighbor rule;;Pseudo nearest neighbor rule;;Pattern classification
To explore the background and basis of the node document
Documents that have the similar content to the node document