SUMMARY
Hoax is fake news, now massively spread through social media. The impact of hoaxes is that people's misperceptions in understanding of news are very high. With the existence of hoaxes are spreading through social media, it requires the public to think smart when receiving the news. Currently, many ways to prevent hoaxes, right now we have Fact Checker Directory Platform which is a truth platform sourced from several fact check sites. On the truth check platform, every news detected as hoaxes has been categorized into specific type of hoax, manually by the validator. For this reason, this research attempts to automatically categorize the types of hoaxes using comparation of Deep Learning with Machine Learning classifications. Deep Learning uses Long Short Term Memory Network (LSTM), while Machine Learning uses Support Vector Machine (SVM) and Multinomial Naive Bayes. Through the build model process, SVM produces the best accuracy quality of 0.74, Multinomial Naïve Bayes produces an accuracy quality of 0.62 while LSTM displays 0.49. The results of low accuracy in LSTM need to be evaluated on model architecture and data normalization during preprocessing.