An Improved KNN Algorithm for Imbalanced Data Based on Local Mean

KNN algorithm is a simple, effective, non-parametric classification, and has been widely used in text classification, pattern recognition, image and spatial classification. Research on improvements about KNN algorithm has broad application prospects and important scientific significance. Based on analysis about classic KNN and its improved algorithms, we find its over-reliance on the choice of k value, large computational overhead, and misclassification in imbalanced data. In order to reduce these deficiencies, we propose an improved KNN algorithm based on local mean. Experimental results indicate that, compared with classic KNN algorithm, the improved KNN has higher accuracy and stability, and has better classification performance in imbalanced data.