Abstract:This paper presents a improved K-NN algorithm. The CURE clustering is carried out to select the subset of the training set. It can reduce the volume of the training set and omit the outlier. Therefore it can lead both to computational efficiency and to higher classification accuracy. In the algorithm, the weights of each feature are learned using neural network. The feature weights are used in the nearest measure computation such that the important features contribute more in the nearest measure. Experiments on several UCI databases and practical data sets show the efficiency of the algorithm.