Nearest Cluster Classifier

Abstract

In this paper, a new classification method that uses a clustering method to reduce the train set of K-Nearest Neighbor (KNN) classifier and also in order to enhance its performance is proposed. The proposed method is called Nearest Cluster Classifier (NCC). Inspiring the traditional K-NN algorithm, the main idea is to classify a test sample according to the tag of its nearest neighbor. First, the train set is clustered into a number of partitions. By obtaining a number of partitions employing several runnings of a simple clustering algorithm, NCC algorithm extracts a large number of clusters out of the partitions. Then, the label of each cluster center produced in the previous step is determined employing the majority vote mechanism between the class labels of the patterns in the cluster. The NCC algorithm iteratively adds a cluster to a pool of the selected clusters that are considered as the train set of the final 1-NN classifier as long as the 1-NN classifier performance over a set of patterns included the train set and the validation set improves. The selected set of the most accurate clusters are considered as the train set of final 1-NN classifier. After that, the class label of a new test sample is determined according to the class label of the nearest cluster center. Computationally, the NCC is about K times faster than KNN. The proposed method is evaluated on some real datasets from UCI repository. Empirical studies show an excellent improvement in terms of both accuracy and time complexity in comparison with KNN classifier.