A Machine Learning Approach for Depression Detection In Sinhala-English Code-Mixed Language

Abstract

The most significant and prevalent mental illness in the world today is depression. Depression is a treatable condition. In Sri Lanka, people would not recognize depression in its earliest stages, when left un-treated contributes to a high rate of suicidal behavior. Most people tend to share depressive feelings and thoughts intentionally or unintentionally on social media. Social media is a viable source of data for researchers engaged in attempts for early detection of depression. There are numerous studies done in the field of depression detection using English language. But depression detection is less researched using code-mixed languages which is a prevalent trend among most social media users. Therefore, the objective of this research is proposing a machine learning approach to detect depression in Sinhala-English code-mixed textual contents posted on social media. Several base and ensemble machine learning algorithms were tested to detect depression using a dataset collected from social media platforms. Evaluating the performance results, it was found that ExtraTreesClassifier outperformed other machine learning algorithms with the highest accuracy of 79.13%, which was then incorporated into the research prototype.