Skip to main content Skip to navigation

Dropout Model Evaluation in MOOCs

Project Overview

The document explores the integration of generative AI in education, particularly its role in enhancing student success through predictive modeling in Massive Open Online Courses (MOOCs). It highlights the challenges faced in predicting student dropout rates and underscores the necessity for robust statistical evaluations of these predictive models. The findings reveal that straightforward clickstream data is more effective in forecasting student attrition compared to more complex features derived from forums and assignments. By leveraging generative AI techniques, educators and institutions can improve engagement and retention strategies, ultimately fostering a more supportive learning environment. The insights gleaned from the analysis advocate for a data-driven approach in educational settings, emphasizing the potential of AI to tailor interventions based on student behavior and engagement patterns, thereby enhancing overall educational outcomes.

Key Applications

Predictive modeling of student dropout in MOOCs

Context: Massive Open Online Courses (MOOCs), targeting instructors and educational institutions

Implementation: Utilized raw clickstream data and database exports from MOOCs to extract features and build predictive models using statistical methods (Friedman and Nemenyi test) to evaluate model performance.

Outcomes: Demonstrated that clickstream-based features significantly outperform forum and assignment-based features in predicting dropout rates.

Challenges: Complex features from forums and assignments underperformed, and there is a lack of consensus on rigorous evaluation methods in the community.

Implementation Barriers

Methodological Barrier

No consensus on methods for rigorous and reproducible statistical inference for evaluating predictive models.

Proposed Solutions: Adopt standardized evaluation methods such as the Friedman and Nemenyi procedure for model selection tasks.

Data Availability Barrier

Complex features are often only available for small subsets of learners, limiting their usefulness.

Proposed Solutions: Focus on feature extraction methods that are generalizable and can be applied to the full learner population.

Project Team

Josh Gardner

Researcher

Christopher Brooks

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Josh Gardner, Christopher Brooks

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies