부산대학교 도서관

Feature selection has become a focus of research in many fields that deal with machine learning and data mining because it makes classifiers cost-effective, faster, and more accurate. In this paper, the impact of feature selection using filter methods such as Information Gain is shown. The impact of feature selection has been analyzed based on the accuracy of two classifiers: J48 and Naïve Bayes. The Airline Customer Satisfaction datasets have been used for comparing with and without applying Information Gain. As a result, J48 achieved 0.33% and 0.29% improvements in accuracy after applying Information Gain for 10-fold and 20-fold cross-validation, respectively compared to Naïve Bayes. Most of the precision and F1-score for J48 with Information Gain have also improved for both evaluation methods compared to Naïve Bayes. In conclusion, J48 seems to be the classifier that is most sensitive to feature selection and has shown improvements compared to Naïve Bayes.