Comparative Analysis of Logistic Regression and XGBoost for Depression Detection from Reddit Posts

Depression is a major mental health disorder that affects 
millions of individuals worldwide and often remains 
undiagnosed due to social stigma and limited access to 
professional care. The increasing use of social media 
platforms such as Reddit provides an opportunity to 
analyze textual expressions that may contain indicators 
of depressive behavior. This study investigates the 
effectiveness of machine learning techniques for 
depression detection using textual data collected from 
Reddit posts. 
A Reddit Depression Dataset containing 7,731 posts was 
analyzed through exploratory data analysis, text 
preprocessing, TF-IDF feature extraction, and VADER 
sentiment analysis. The extracted features were 
evaluated using two machine learning classifiers: 
Logistic Regression and XGBoost. Performance was 
assessed using accuracy, precision, recall, F1-score, 
confusion matrix analysis, and ROC-AUC. 
Experimental results demonstrated strong classification 
performance. Logistic Regression achieved an accuracy 
of 94.44% and an AUC of 0.986, while XGBoost 
achieved a slightly higher accuracy of 94.70%. The 
findings indicate that TF-IDF lexical features provide 
substantial predictive information for distinguishing 
depressive and non-depressive posts. Sentiment analysis 
further revealed noticeable differences in emotional 
polarity between the two classes. 
The study presents a reproducible and computationally 
efficient framework for depression detection using 
publicly available data and open-source tools. The 
proposed workflow is suitable for academic research 
and educational environments where interpretability, 
simplicity, 
and reproducibility are important 
considerations.