Random Forest Exceeds Support Vector Machine In Election Abstention Sentiment Analysis

Abstract

General Background Public opinions regarding political participation are increasingly expressed on social media platforms. Specific Background During the 2024 Indonesian voting period, the white group phenomenon generated significant unstructured textual data on the X platform. Knowledge Gap Existing literature presents competitive capabilities between classification models, necessitating a targeted evaluation of their performance specifically on political non-voting discourse. Aims This study compares the classification capabilities of two predictive algorithms regarding public views on the 2024 white group movement. Results Utilizing 1,008 textual data points processed via term frequency-inverse document frequency, the decision tree-based model achieved higher metrics with 75% accuracy, 75% precision, 75% recall, and a 73% F1-score. Conversely, the hyperplane-based algorithm recorded 72% accuracy, 69% precision, 72% recall, and a 69% F1-score. Furthermore, data distribution revealed that 48.31% supported non-voting, 35.91% opposed it, and 15.78% remained neutral. Novelty The research uniquely integrates a Streamlit application that visually maps the entire machine learning pipeline alongside political distribution patterns. Implications These analytical findings provide institutions with critical insights to formulate targeted public communication strategies and encourage future citizen participation. Highlights Predictive modeling demonstrated a 75% accuracy rate for identifying unstructured social media text patterns. Non-voting advocacy dominated the digital public discourse with 487 documented instances during the observation period. An interactive web application was successfully deployed to visualize the entire data processing workflow. Keywords Machine Learning; Text Classification; Public Opinion; Feature Extraction; Data Visualization