A Machine Learning based Depression Analysis and Suicidal Ideation Detection System using Questionnaires and Twitter

Depression as a disorder has been a great concern in our society and has been perpetually a hot topic for researchers in the world. Despite the massive quantity of analysis on understanding individual moods together with depression, anxiety, and stress supported activity logs collected by pervasive computing devices like smartphones, foretelling depressed moods continues to be an open question. In this paper, we have proposed a depression analysis and suicidal ideation detection system, for predicting the suicidal acts based on the level of depression. We collected real time data from students and parents by making them fill questionnaires similar to PHQ-9 (Parent health questionnaire) consisting of questions like What's your age? or Are you regular in school/college? and processed it into meaningful data with related features like age, sex, regularity in the school, etc. Then, classification machine algorithms are used to train and classify it in five stages of depression depending on severity - Minimal or none, mild, moderate, moderately severe and severe. Maximum accuracy i.e. 83.87 % was achieved by using XGBoost classifier in this dataset. Also, data was collected in the form of tweets and were classified into whether the person who tweeted is in depression or not using classification algorithms. Logistic Regression classifier gave the maximum accuracy i.e. 86.45 % for the same.