Extensions of Local-Mean K-Nearest Neighbor Classifier with Various Distance Metrics

Abstract

K-Nearest Neighbor (KNN) classification algorithm is one of the simplest methods in data mining classification technology. The LMKNN algorithm based on this method uses the local average vector of each category to classify the data. However, because distance measurement only uses Euclidean distance algorithm, it is not suitable for some cases. Therefore, this paper proposes a local mean k-nearest neighbor classification algorithm based on different distance algorithms. And compare the classification capabilities of the LMKNN model based on four distance measurement algorithms through experiments. The results show that the LMKNN algorithm based on Minkowski and City block distance metric schemes has more advantages than the LMKNN algorithm based on Euclidean distance metric schemes.