MFZ-KNN — A modified fuzzy based K nearest neighbor algorithm

KNN is amongst the simplest top ten classification algorithm of data mining. Being effective and efficient it has some drawbacks which cannot be overlooked. Moreover, real world data is fuzzy in nature. To overcome this drawback fuzzy KNN was introduced which was based on fuzzy membership. But, it had large time complexity as the membership is calculated at the classification period. To improve this, we have proposed a modified fuzzy based KNN algorithm MFZ-KNN whereby fuzzy clusters are obtained at preprocessing step and the membership of the training data set is computed in reference with the centroid of the clusters. This reduces the complexity of time remarkably. We have implemented the algorithm in MatLAB and Netbeans IDE using standard UCI data set-Wine. The results prove that it is better than both conventional KNN and fuzzy KNN in terms of accuracy and time.