Modified Weighted K-Nearest Neighbor

KNN is one of the most popular classification algorithms which are widely used by researchers. KNN is simple, effective, nonparametric, and has a good performance especiallywhen there are no explicit knowledge of data. Of course the traditional KNN algorithms has some limitations such as: i) theperformance is solely dependent on training set and uneven density of training data can lead to reduction of accuracy rate of classification, ii) all K nearest neighbors is equal and there is noweight difference between them, iii) the value of K has critical effect on the algorithm performance. In this paper a new modifiedweighted algorithm for KNN classification (MWKNN) is presented which by evaluating training set, determine each samplestatus on the training set and then use this information for classifying new data. Experimental results indicate that ourmethod could achieve better or at least comparable performancewith traditional KNN and also some other KNN improved algorithms.

Modified Weighted K-Nearest Neighbor Keywords:

KNN , classification , Weighted K-nearest neighbor , pattern recognition