A Comparative Study of Machine Learning Models for Stress Detection in Academic Social Media Posts

Key Points

• The aim is to evaluate various machine learning models for detecting stress in academic social media posts.
• Utilized the Dreaddit dataset for analysis.
• Assessed five different classifiers across multiple feature configurations.
• Developed a weighted ensemble model using VADER sentiment features.
• Achieved 77.90% accuracy in stress detection.
• Obtained a 79.48% F1-Score on 715 test posts.
• Demonstrated effective performance of ensemble models in social media text classification.

Abstract