Evaluation of Feature Extraction TF-IDF in Indonesian Hoax News Classification

Hoax is information that generally does not exist or not true and dangerous because it can mislead people perception and give an assumption to the news that not truly happen become a truth. In 2016, the election for US president has been held, due to this event, there are several fake news spread through social media. This incident has to take responsibility for the huge effect on the election result. Therefore, a method for detecting hoax should be developed to prevent the impact of the hoax. The purpose of this research is evaluating several machine learning methods for detecting fake news. The dataset that uses in this research is hoax news collected from Department Communication and Information of Central Java Province (November 20018 and December 2018) and www.turnbackhoax.id (December 2018 until March 2019). The number of documents used for this research is 150 documents consist of 75 documents with true category and 75 documents hoax category. Feature extraction used in this research is TF-IDF algorithm. This research evaluates the performance of several machine learning methods such as MNB, k-NN, Logistic classification, and LSVM in classifying fake news. Base on the test result using 10-fold validation, Logistic classification have the best accuracy with 84.67%.