In this paper, we develop a novel Distance-weighted k-nearest Neighbor rule (DWKNN), using the dual distance-weighted function. The proposed DWKNN is motivated by the sensitivity problem of the selection of the neighborhood size k that exists in k-nearest Neighbor rule (KNN), with the aim of improving classification performance. The experiment results on twelve real data sets demonstrate that our proposed classifier is robust to different choices of k to some degree, and yields good performance with a larger optimal k, compared to the other state-of-art KNN-based methods.
|Number of pages||8|
|Journal||Journal of Information and Computational Science|
|Publication status||Published - Jun 2012|
- K-nearest neighbor rule
- Pattern classification
- Weighted voting