A Review Paper: Accuracy of Machine Learning for Depression Detection in Social Media

Abstract

There are so many health problems that affect humans. One of them is depression. Depression is a mental health disorder that would trigger suicidal tendencies if not treated carefully. People who are depressed tend to have less concentration and productivity. However, detecting depression is not easy due to the self-denial of some patients, and they keep depression untreated and undiagnosed. Some factors of untreated or undiagnosed depression are poor knowledge and recognition in many places the patient is shy to talk to a psychologist, and the stereotypes in public that say people who come to a psychologist are “insane.” Depression symptoms of a user can be shown in social media posts, and these symptoms can be detected using a machine learning algorithm. These Machine learning algorithms can be an alternative for detecting depression or as a supporting document for psychologist diagnoses. The algorithm obtains accurate that varies depending on the dataset. For this reason, we conducted a systematic literature review to find out which machine learning has the best accuracy in detecting depression. We also provide information about stable algorithms to detect a given dataset and the popular dataset used in previous studies based on the most frequent text that is easy to test. In conclusion, the greatest accuracy is obtained from Logistic Regression with an accuracy value of 99.80%. Stable algorithms are obtained by LR and SVM because the machine learning method obtains values above 70%. The most popular dataset used in previous studies is the Twitter dataset.