Depression is one of the most common health problems which has a significant impact for its victims. Depression is characterized or influenced by many aspects of life, including life experiences, work, and social life. In 2018, approximately 6.1% from 267.7 million people experienced mental disorders in Indonesia. The stigma of psychiatric illness and low awareness of undergoing treatment to experts is becoming the underlying factor for such percentages.
Nowadays, self-expression is often done via social media posts or comments. Twitter is one of the tools for self-expression in textual form. In this research, we collected the text dataset which contain keywords related to depression from Twitter and do a laborious process involving a psychiatrist for labelling each data to two classes, namely “indicated depression” or “not indicated”. Based on the labelled dataset, we build a predictive model by using Multinomial Naïve Bayes (MNB) and Complement Naïve Bayes (CNB) as a classifier and Term Frequency - Inverse Document Frequency (TF–IDF) as feature extractor. From our experiments, the combination of TF-IDF and MNB are able to achieve the best F-score of 91.30% while the TF- IDF and CNB are able to achieve the best F-score of 91.98%.