Penerapan SVM dengan Seleksi Fitur Mutual Information untuk Memprediksi Sentimen PEMILU 2024

Abstract

A wealth of information on the 2024 Indonesian Election floods Twitter, from campaign schedules to candidate profiles and the latest survey results indicating candidate popularity. This information overload poses challenges in discerning comments' sentiment. Manual classification is feasible but time-consuming. Hence, this study aims to streamline data analysis for the 2024 Election. It employs a dataset of 1000 entries categorized as positive or negative. Support Vector Machine (SVM) with Mutual Information feature selection is utilized for classification. Results reveal that Mutual Information feature selection enhances SVM performance. Without it, SVM achieves 88% accuracy and 87.9% f-measure using the rbf kernel (C=1, ?=2), computed in about 0.07 seconds. With feature selection, SVM's accuracy improves to 90%, and f-measure to 89.9% with 60% features, using rbf kernel (C=10, ?=0.5), reducing computation time to 0.02 seconds, optimizing both performance and efficiency. The website system scored 88.63 in usability, higher than the global average of 68, based on a SUS questionnaire with 10 questions and 20 respondents. This indicates excellent performance and user satisfaction, as evaluated from the web system.