COMPARISON OF FEATURE SELECTION BASED ON COMPUTATION TIME AND CLASSIFICATION ACCURACY USING SUPPORT VECTOR MACHINE

Abstract

The goal of this research to compare Chi-Square feature selection with Mutual Information feature selection based on computation time and classification accuracy. In this research, people's comments on Twitter are classified based on positive, negative, and neutral sentiments using the Support Vector Machine method. Sentiment classification has the disadvantage that it has many features that are used, therefore feature selection is needed to optimize a sentiment classification performance. Chi-square feature selection and mutual information feature selection are feature selections that both can improve the accuracy of sentiment classification. How to collect the data on twitter taken using the IDE application from python. The results of this study indicate that sentiment classification using Chi-Square feature selection produces a computation time of 0.4375 seconds with an accuracy of 78% while sentiment classification using Mutual Information feature selection produces an accuracy of 80% with a required computation time of 252.75 seconds. So that the conclusion are obtained based on the computational time aspect, the Chi-Square feature selection is superior to the Mutual Information feature selection, while based on the classification accuracy aspect, the Mutual Information feature selection is more accurate than the Chi-Square feature selection. The recommendations for further research can use mutual information feature selection to get high accuracy results on sentiment classification