Increasing the speed of fuzzy k‐nearest neighbours algorithm

Abstract

AbstractFuzzy k‐nearest neighbour (FKNN) is one of the most convenient classification approaches. The main challenge of this method is associated with finding the optimal values of its two hyperparameters. The present study attempts to decrease the running time of this approach by reducing the number of its hyperparameters through omitting the hyperparameter k. In the training phase of FKNN approach, the membership degree of each training data is refined by crisp KNN voting whereas a fuzzy voting is used in the training phase of our proposed approach. Training and test phases time complexities of our proposed approach are better than those of FKNN approach. The experiments on real data sets indicate that the accuracy of our proposed approach called ultra FKNN is higher than FKNN approach due to applying fuzzy voting instead of crisp voting in training phase. In addition, the training, test, and running time of our proposed approach are considerably less than those of FKNN.