A Method Based on the Edited FKNN by the Threshold Value

Abstract

The edited technique is of great importance in pattern recognition. The classical edited fuzzy technique use fuzzy k nearest neighbors (FKNN) to take out some useless samples which was classified erroneously in the editing process. In this paper, a proposed edited fuzzy k nearest neighbors based on threshold is developed, which not only consider the maximum membership value but also consider that whether the maximum value is bigger than the given threshold value. We use refer samples to classifier the test samples by FKNN, in which we not only select samples classified correctly but also consider the maximum membership value. That is, threshold value is used to take out some samples that was unreliable. Several comparisons are made between the proposed edited FKNN and the classical edited FKNN, which shown that the proposed method is better then the classical method.