A new method of learning weighted similarity function to improve predictions of Nearest Neighbor rule

The performance of Nearest Neighbor (NN) classifier is highly dependant on the distance (or similarity) function used to find the NN of an input test pattern. In order to optimize the accuracy of the NN ride, a weighted similarity function is proposed. In this scheme, a weight is assigned to each training instance. The weights of training instances are used in the generalization phase to find the NN of an input test pattern. To specify the weights of training instances, we propose a learning algorithm that attempts to minimize the leave-one-out (LV1) error rate of the classifier on train data. The proposed approach is assessed using a number of data sets from UCI corpora. Simulation results show that the proposed method improves the generalization accuracy of the basic NN and results are comparable to or better than other methods proposed in the past to learn the distance function.