A Comparative Analysis of Balancing Techniques and Attribute Reduction Algorithms

摘要

In this study we analyze several data balancing techniques and attribute reduction algorithms and their impact over the information retrieval process. Specifically, we study its performance when used in biomedical text classification using Support Vector Machines (SVMs) based on Linear, Radial, Polynomial and Sigmoid kernels. From experiments on the TREC Genomics 2005 biomedical text public corpus we conclude that these techniques are necessary to improve the classification process. Kernels get some improvements about their results when attribute reduction algorithms were used. Moreover, if balancing techniques and attribute reduction algorithms are applied, results obtained with oversampling are better than subsam-pling.