Improved k nearest neighbors Transductive Confidence Machine for pattern recognition

Abstract

Transductive Confidence Machine(TCM) is an effective machine- learning algorithm. But its classification results are not satisfying under high confidence level. Therefor an improved algorithm, named TCM-IKNN, is put forward by means of improving strangeness measure method on the basis of traditional TCM-KNN. The results of the experiment on parts of UCI dataset show that the TCM-IKNN algorithm using the improved strangeness measure can increase the correct rate of predictions, reduce the number of uncertain predictions in both online and offline learning settings, be superior to traditional TCM-KNN.

3 Figures and Tables

Cite this paper

@article{Lilin2010ImprovedKN, title={Improved k nearest neighbors Transductive Confidence Machine for pattern recognition}, author={Cui Li-lin and Zhu Hai-chao and Zhang Lin-ke and Luan Rui-peng}, journal={2010 International Conference On Computer Design and Applications}, year={2010}, volume={3}, pages={V3-172-V3-176} }