<![CDATA[The spread of hoax news in Indonesia has become increasingly massive in the digital era, with the Ministry ofCommunication and Digital Technology (Komdigi) recording 1,923 hoax content pieces in 2024. Manual verification isineffective in addressing the continuously growing volume of information, while existing automatic detection systems still have limitations in accuracy and real-world implementation. This research aims to build an Indonesian hoax news classification model using the Multinomial Naive Bayes method based on TF-IDF, analyze the impact of TF-IDF on text feature representation, and measure model performance based on accuracy, precision, recall, and F1-score metrics. Data was collected through web scraping from seven online news portals, producing 59,795 news headlines with balanced distribution between the hoax class (47.53%) and non-hoax class (52.47%). Text preprocessing stages include case folding, stopword removal, and stemming using the Enhanced Confix Stripping (ECS) algorithm from the PySastrawi library. Features were extracted using TF-IDF with ngram_range=(1,2) and sublinear_tf=True configuration, producing 5,000 representative features. The dataset was split in an 80:20 ratio and optimized using GridSearchCV with 5-fold crossvalidation. The baseline model evaluation achieved accuracy of 94.00%, precision of 93.53%, recall of 93.86%, and F1-score of 93.69%, with balanced performance across both classes. After hyperparameter optimization (alpha=5.0), the crossvalidation score increased to 94.18% with hoax class precision reaching 95%. Testing on new data yielded 85.7% accuracy, demonstrating the system's effectiveness as an automated hoax detection tool for the Indonesian language]]>
Copyrights © 2026
