A Relative Weight-Based Fuzzy k-Nearest Neighbour Classifier for Chemical Analysis of Wine and Glass Identification

The nearest neighbour (NN) classifier is a widely used classifier that classifies a test pattern based on the class of its closest neighbour. The NN classifier has several variants, namely k-NN, modified k-NN (Mk-NN) and fuzzy k-NN (Fk-NN). These variants classify a test pattern using some or all of the training patterns. The k-NN classifier is different from the Mk-NN classifier in terms of weight, whereas the Mk-NN classifier is different from the Fk-NN classifier in terms of membership value. This paper introduces a new classifier called relative weight-based Fk-NN (RWFk-NN) by taking advantage of Mk-NN and Fk-NN classifiers. It also presents a comparative analysis of these classifiers using three techniques, namely k-fold cross-validation, k-test analysis and m-test analysis, on the wine and glass identification datasets from the University of California Irvine (UCI) machine learning repository. The performance of all the classifiers is measured in terms of accuracy. The findings reveal that the RWFk-NN classifier outperforms all the other classifiers in both datasets.