| 520 |
3 |
|
|a In nearest neighbor classification, fuzzy sets can be used to model the degree of membership of each instance to the classes of the problem. Although the fuzzy memberships can be set by analyzing local data around each instance, there may be still a lack of knowledge associated with the assignation of a single value to the membership. This is caused by the requirement of determining in advance two fixed parameters: k, in the definition of the initial membership values and m, in the computation of the votes of neighbors. Thus, the two fixed parameters only allow the flexibility of membership using a single value only. To overcome this drawback, a new approach of interval valued fuzzy sets k-nearest neighbors (IVFKNN) that incorporating interval valued fuzzy sets for computing the membership of instance is presented that allows membership values to be defined using a lower bound and an upper bound with the length of interval. The intervals concept is introduced to assign membership for each instance in training set and represents membership as an array of intervals. The intervals also considered the computation of the votes with the length of interval. In order to assess the classification performance of the IVFKNN classifier, it is compared with the competing classifiers, such as k-nearest neighbors (KNN) and fuzzy k-nearest neighbors (FKNN), in terms of the classification accuracy on publicly available Finger Vein USM (FV-USM) image database which was collected from 123 volunteers. The experimental results remark the strong performance of IVFKNN compared with the competing classifiers and show the best improvement in classification accuracy in all cases.
|
