<![CDATA[The rapid advancement of information technology in Indonesia has led to a massive spread of digital disinformation, commonly known as an infodemic. The inability to filter inaccurate information manually necessitates a reliable, automated hoax detection system. This study aims to implement and evaluate the Multinomial Naïve Bayes algorithm combined with Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction to classify news articles as either factual or hoax. The research utilizes a dataset of 2,910 Indonesian news articles published in 2025, collected from verified national news portals and fact-checking websites. The text data underwent comprehensive preprocessing—including case folding, cleansing, stopword removal, and stemming—before being evaluated using 5-Fold Cross-Validation and an 80:20 data split. Experimental results demonstrate that the Naïve Bayes model achieves highly stable and competitive performance, recording an accuracy of 93.81%, a precision of 93.84%, a recall of 93.81%, an F1-Score of 93.82%, and a 5-Fold Cross-Validation F1-Score of 93.39%. Notably, the algorithm exhibited a significantly low False Negative rate, missing only 15 hoax documents out of 582 test samples. Furthermore, the trained model was successfully integrated into a real-time, web-based user interface using Streamlit. This practical implementation provides an accessible and efficient initial screening tool for the general public and journalists to assist in verifying news authenticity, thereby supporting efforts to mitigate the impact of digital hoaxes.]]>
Copyrights © 2026
