Comparison of Support Vector Machine and Extreme Learning Machine for Sentiment Analysis of Free Nutritious Meal Program on X Application

Abstract

In the current digital era, data analysis has become a critical field, particularly in processing unstructured text data from online interactions. A prime example is public sentiment analysis on platforms like Application X, which is valuable for evaluating government policy effectiveness. However, the diversity of opinions and the unstructured nature of text data pose challenges in achieving accurate classification. This study aims to implement and compare the performance of Support Vector Machine (SVM)—known for its effectiveness in high-dimensional data—and Extreme Learning Machine (ELM)—renowned for computational efficiency—in sentiment classification. The dataset consists of tweets from the X social media platform regarding the Free Nutritious Meal Program. We evaluated four SVM kernel types and four ELM activation functions. Results demonstrate that the linear-kernel SVM achieved the highest accuracy (88.4%) with an F1-gap of 5.82%, while the linear-activation ELM attained 83.8% accuracy with a 2.54% F1-gap. These findings suggest that linear-kernel SVM is more suitable for sentiment analysis of similar datasets. Moreover, the sentiment analysis revealed a predominance of negative sentiment, reflecting public concern and criticism regarding the program's implementation.