Improved k-NN Algorithm for Text Classification

Abstract- Over the last twenty years, text classification has become one of the key techniques for organizing electronic information such as text and web documents. The k-Nearest Neighbor (k-NN) algorithm is a very well known and popular algorithm for text classification. The k-NN algorithm determines the classification of new document by the class of its k-nearest neighbor. In this paper we propose an improved k-NN algorithm with a built-in technique to skip a document from training corpus without looking inside the document if it is not important, which improves the performance of the algorithm. It also has an improved decision rule to identify class from k-nearest neighbor to improve the accuracy by avoiding bias of dominating class with large number of documents. We conduct experiments on benchmark text classification datasets. The new and improved k-NN algorithm is suitable for other applications as well