Identifying Fake News in Indonesian via Supervised Binary Text Classification

Fake news detection has gained growing interest from both the industry and research community all around the world, including Indonesia. Based on recent surveys, people could receive fake news daily, if not more than once. The research community and practitioners, supported by the government, are trying to fight back the spreading of fake news. This paper aims to implement a supervised machine learning approach using the Multi-Layer Perceptron (MLP) for classifying news article in order to detect fake news articles and differentiate them from the valid ones, via a binary text classification approach. Furthermore, this paper uses TF-IDF in comparison with the Bag of Words model to extract features along with the use of the n-gram model. Based on the result, our final model could achieve a hoax precision and recall score of 0.84 and 0.73, respectively, and a macro-averaged F1-score of 0.82. Furthermore, our paper shows that some preprocessing methods such as stemming and stop-word removal could be very time-consuming while only barely affecting the performance of our classifier model using the dataset in this research for identifying fake news.