Mental Illness Detection in Social Media Data Using Sub-emotions Model and Ensemble-Based Hybrid Technique

Depression is a psychological expression consisting of low mood, loneliness, sadness, anger and emptiness that disrupt relationships, jobs and the health of an individual's daily life. In severe cases, it leads to suicide attempts. With the rise of social media usage for sharing thoughts and interpersonal communications, these platforms have become powerful tools for self-expression. Several studies have demonstrated the detection of early signs of depression using social media data, leveraging various Machine Learning (ML) algorithms and Natural Language Processing (NLP) techniques. One promising approach we built in our previous work is the Bag of Sub-Emotions (BoSE) method, which extracts sub-emotions from textual data, allowing fine-grained detail of the emotions expressed. Additionally, ML algorithms play a vital role in detecting depression content and categorizing it into normal or depressed. The goal of this paper is to find depression in social media posts by using the BoSE model with machine learning algorithms. We begin with employing the BoSE model developed in our previous work for sub-emotion generation. This sub-emotional dataset is selected to train Support Vector Machine (SVM), Random Forests (RF), Decision Trees (DT), Multinomial Naive Bayes (MNB), Logistic Regression (LR), AdaBoost with SVM, XGBoost and Convolutional Neural Network (CNN) models. The performance of these models is evaluated and found that the CNN model provides better accuracy compared with other models.