Skip to main content Skip to navigation

Modeling the EdNet Dataset with Logistic Regression

Project Overview

This document explores the application of generative AI in education, particularly through the use of logistic regression models in the EdNet Kaggle competition for student modeling, underscoring the challenges posed by the opaque nature of neural networks that complicate understanding learner behavior. It emphasizes the necessity for transparent learner model predictions to inform pedagogical decisions effectively. The authors share their development of the LKT (Logistic Knowledge Tracing) tool, designed to facilitate the creation of logistic regression models that enhance learning predictions. Additionally, the document highlights the role of pedagogical decision rules (PDRs) in maximizing the practical utility of these learner models. By integrating these insights, the findings suggest that comprehensible AI-driven approaches can significantly improve the educational landscape, enabling better-targeted interventions and supporting educators in making informed decisions based on data-driven insights.

Key Applications

LKT (Logistic Knowledge Tracing)

Context: Used in the EdNet Kaggle competition for student modeling

Implementation: Developed an R package to create logistic regression-based models of student learning

Outcomes: Achieved a high AUC score for model predictions and facilitated understanding of student learning features

Challenges: Lack of detailed data and features made model creation more difficult; reliance on static decision rules limited the model's effectiveness

Implementation Barriers

Data Limitations

Missing data details and item descriptions impeded the model development process

Proposed Solutions: Encouraging better data transparency and providing comprehensive datasets for model training

Model Complexity

The complexity of the models used (e.g., deep learning) may create a disconnect for learning scientists unfamiliar with these methods

Proposed Solutions: Integrating features from simpler models into complex models to make them more comprehensible for learning scientists

Project Team

Philip I. Pavlik Jr

Researcher

Luke G. Eglington

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Philip I. Pavlik Jr, Luke G. Eglington

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