Hoax News Detection on Urban Issues Using Naive Bayes with Chi-Square Feature Selection

Abstract

The rapid spread of hoax news in digital media has become a serious concern because it can mislead the public, influence opinions, and potentially trigger social unrest, particularly in urban-related issues. The increasing volume of online information also makes manual verification difficult, highlighting the need for automatic hoax detection systems using machine learning approaches. This study aims to develop a hoax news detection model using the Multinomial Naive Bayes algorithm combined with Chi-Square feature selection to improve classification efficiency. The dataset consists of 1,000 Indonesian news articles related to urban issues, consisting of hoax and factual news collected through web scraping from fact-checking websites and online news portals. The research process includes text preprocessing stages such as case folding, tokenizing, stopword removal, and stemming, followed by TF-IDF feature extraction. Chi-Square feature selection was applied to reduce feature dimensionality by selecting the most relevant terms. The dataset was divided into training and testing sets using an 80:20 ratio. Model performance was evaluated using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The experimental results show that the Multinomial Naive Bayes model without feature selection achieved the highest accuracy of 98.5%. Meanwhile, the application of Chi-Square feature selection with 200 selected features achieved an accuracy of 90.5%, while using 100 features resulted in an accuracy of 86.5%. These results indicate that although feature selection slightly reduces classification performance, it significantly reduces feature dimensionality while maintaining acceptable performance. Overall, this study demonstrates that Multinomial Naive Bayes is effective for hoax news classification, while Chi-Square feature selection provides an efficient alternative when dimensionality reduction is required.