|
|
|
| A Novel Template Reduction K-Nearest Neighbor Classification Method Based on Weighted Distance |
| Yang Jin-fu Song Min Li Ming-ai |
| College of Electronic Information and Control Engineering, Beijing University of Technology, Beijing 100124, China |
|
|
|
|
Abstract As a nonparametric classification algorithm, K-Nearest Neighbor (KNN) is very efficient and can be easily realized. However, the traditional KNN suggests that the contributions of all K nearest neighbors are equal, which makes it easy to be disturbed by noises. Meanwhile, for large data sets, the computational demands for classifying patterns using KNN can be prohibitive. In this paper, a new Template reduction KNN algorithm based on Weighted distance (TWKNN) is proposed. Firstly, the points that are far away from the classification boundary are dropped by the template reduction technique. Then, in the process of classification, the K nearest neighbors’ weights of the test sample are set according to the Euclidean distance metric, which can enhance the robustness of the algorithm. Experimental results show that the proposed approach effectively reduces the number of training samples while maintaining the same level of classification accuracy as the traditional KNN.
|
|
Received: 18 January 2011
|
|
|
|
Corresponding Authors:
Song Min
E-mail: songmin@emails.bjut.edu.cn
|
|
|
|
|
|
|
2011, Vol. 33 
