Depression Detection From Social Media Textual Data Using Natural Language Processing and Machine Learning Techniques

Depression, a widespread mental health condition with substantial personal and societal impacts, necessitates early detection for effective intervention. The focus of this thesis is crafting a dependable and precise system to identify depression from social media text, employing natural language processing (NLP) and machine learning (ML). Drawing from diverse social media posts, both from individuals with and without depression, the primary aim is to preprocess this textual data effectively. NLP techniques, including tokenization, stemming, N-gram, Countvectorizer analysis, and TF-IDF, convert raw content into meaningful representations capturing linguistic and emotional facets of depression. Utilizing the preprocessed data, machine learning algorithms acquire discriminating patterns through feature extraction. Diverse ML techniques—Stochastic Gradient Descent (SGD), Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), Support Vector Machines (SVM), KNearest Neighbor (KNN), Multi-Layer Perceptron (MLP), etc.—are employed. Trained on annotated data and extracted attributes, these algorithms discern depressive from nondepressive social media posts. Model performance assessment incorporates metrics like accuracy, precision, recall, and F1score. Results underscore the efficacy of this approach in detecting depression from social media text. The system consistently demonstrates high accuracy and performance across various evaluation criteria. This thesis concludes by thoroughly exploring depression diagnosis via NLP and ML algorithms applied to social media content.