Weighted Version of Extended Nearest Neighbors

摘要

In K-nearest neighbor method (KNN), the test data is assigned to a class with the highest frequency in its nearest neighbors. One of KNN challenges is its insufficient accuracy especially for small or imbalance dataset. Recently, to tackle this challenge, extended nearest neighbor method (ENN) was proposed. ENN considers not only who the nearest neighbors of the test data are, but also who consider the test data as their nearest neighbors. KNN and ENN use hard voting to determine test data class label. In this paper, an ENN-based algorithm is proposed which uses soft or weighted voting to determine test data class label. Unlike ENN, our proposed method has no training phase, and test phase time complexity of our proposed method is better than ENN. Experiments on the real dataset show that the accuracy, test time and training time of our proposed method is better than those of ENN.