An Improved kNN Based on Class Contribution and Feature Weighting

• Published in: International Conference on Measuring Technology and Mechatronics Automation (Print) pp. 313 - 316
• Main Authors: Huang, Jie, Wei, Yongqing, Yi, Jing, Liu, Mengdi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.02.2018
• Subjects: class contribution; Classification algorithms; Computer science; Data mining; Euclidean distance; feature weighting; kNN; Liver; Training
• DOI: 10.1109/ICMTMA.2018.00083

Aiming at the problem that the kNN algorithm is susceptible to the choice of k-nearest neighbors and the method of class judgment, this paper propose a kNN algorithm based on class contribution and feature weighting called DCT-kNN. Firstly, using traditional kNN to calculate accuarcy of original dataset and of the data lack of each dimension feature successively. Then by comparing two accuarcies to weight the feature and to calculate the weighted distance, by which the k-nearest neighbors are obtained. Finally, by using class contribution which combines the number of k-nearest neighbors and their mean distance, the final labels of the samples are obtained. The comparison experiment of UCI datasets showed a certain degree of improvement in classification accuracy of the proposed method.