K-NN: ESTIMATING AN ADEQUATE VALUE FOR PARAMETER K

The k-NN (k Nearest Neighbours) classification technique is characterized by its simplicity and efficient performance on many databases. However, the good performance of this method relies on the choice of an appropriate value for the input parameter k. In this work, we propose methods to estimate an adequate value for parameter k for any given database. Experimental results have shown that, in terms of predictive accuracy, k-NN using the estimated value for k usually outperforms k-NN with the values commonly used for k, as well as well-known methods such as decision trees and naive Bayes classification.