Using optimally weighted fussy K-nearest neighbor in classification of speech and pattern data sets

Nowadays, information technology (IT) is widely used in many applications in the real world. For example KNN method is highly used in pattern and speech recognition systems in order to classification of samples. This paper compares simple KNN with Fuzzy KNN and Optimally Weighted Fuzzy KNN. We use them in order to classification of speech and pattern databases. The k-nearest neighbor rules assigns crisp memberships of samples to class labels; whereas, the Fuzzy KNN rule replaces crisp memberships with fuzzy memberships. On the other hand, membership assignment by the Fuzzy KNN algorithm has a disadvantage in that it depends on the choice of some distance function, which is not based on any principle of optimality. To overcome this problem, a computational scheme is introduced for determining optimal weights to be combined with different fuzzy membership grades for classification by the fuzzy KNN approach which is called Optimally Weighted Fuzzy KNN. Experimental results show that the Optimally Weighted Fuzzy KNN method outperforms other two methods in variety of data sets. In the UCI Glass data set we improve the classification accuracy up to 4.45 % compare to KNN and 2.11 compare to Fuzzy KNN.