Performance Comparison and Optimization of Text Document Classification using k-NN and Naïve Bayes Classification Techniques

Abstract

In the current era, information is available in several different formats, such as text, image, video, audio and others. Corpus is a collection of documents in a large volume. By using Information Retrieval (IR), it is possible to obtain an unstructured information and automatic summary, classification and clustering. This research is to focus on data classification using two out of the six approaches of data classification, which is k-NN (k-Nearest Neighbors) and Naïve Bayes. The text documents used is in XML format. The Corpus used in this research is downloaded from TREC Legal Track with a total of more than three thousand text documents and over twenty types of classifications. Out of the twenty types of classifications, six are chosen with the most number of text documents. The data is processed using RapidMiner software and the result shows that the optimum value for k in k-NN occurs at k=13. Using this value for k, the accruacy in average reached 55.17 percent, which is better than using Naïve Bayes which is 39.01 percent.