A Classifier Combining Local Distance Mean and Centroid for Imbalanced Datasets

Abstract

The K-Nearest Neighbor (KNN) algorithm is widely used in practical life because of its simplicity and easy understanding. However, the traditional KNN algorithm has some shortcomings. It only considers the number of samples of different classes in k neighbors, but ignores the distance and location distribution of the unknown sample relative to the k nearest training samples. Moreover, classes imbalance problem is always a challenge faced with the KNN algorithm. To solve the above problems, we propose an improved KNN classification method for classes imbalanced datasets based on local distance mean and centroid (LDMC-KNN) in this paper. In the proposed scheme, different numbers of nearest neighbor training samples are selected from each class, and the unknown sample is classified according to the distance and position of these nearest training samples. Experiments are performed on the UCI datasets. The results show that the proposed algorithm has strong competitiveness and is always far superior to KNN algorithm and its variants.