Detecting Teaching Presence in Online Discussions Using Tree-Based Classifiers

• In Event: Elements of Instructional Design in Online Settings

Abstract

This study uses five tree-based machine learning models to explore the automatic classification of teaching presence (TP) in online discussions. Drawing on the Community of Inquiry (CoI) framework, we analyzed 349 posts from two graduate-level courses and extracted 216 linguistic features. Models were evaluated using nested cross-validation and three metrics: F1 score, Cohen’s Kappa, and AUC. Results show that Gradient Boost and XGBoost performed best, with F1 scores above 0.89 and AUC values of 0.897 for Facilitating Discourse and 0.926 for Direct Instruction. These findings suggest that machine learning can effectively detect TP and support instructional analysis. Future work should examine model generalizability and the alignment between feature importance and expert knowledge.

Authors

• Huan Kuang, Florida State University

• Secil Caskurlu, Florida State University

• Hui Shi, Florida State University

• Kadir Kozan, Florida State University

• Seungjie Clarice Han, Florida State University

• Sihan Jian, Florida State University

• Jaesung Hur, Florida State University