Electronic Theses and Dissertations

Eric Hicks

Date

2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Computer Science

Committee Chair

Vinhthuy Phan

Committee Member

Brandon Booth

Committee Member

Deepak Venugopal

Committee Member

Xiaolei Huang

Abstract

CS Education has been plagued with high fail and dropout rates for some time. This has put a larger strain on traditional teaching practices and lead to many attempts to assist struggling students. To deal with these problems, many teachers started to incorporate software tools into their curriculum. By using platforms that came with copious examples and built in testing features, students were able to practice and learn on their own time with immediate feedback, although at a lesser quality. Researchers have often used student data to obtain a better understanding of how students will perform in the future with the goal of leveraging this knowledge to help students better perform. However, most of this work has been being focused on final grade predictions based off of homework grades, which means students have often already completed more as yet ungraded work, thereby compounding errors further. By focusing on smaller, more frequent coding exercises, predictions can be made in time for interventions before the student has moved on. I demonstrated the viability of using using in-class coding to improve student performance and participation, verified other experiments on the use of Active Learning techniques in classrooms, worked on determining the outcomes associated with giving students multiple attempts, and successfully modeled student performance on semi-weekly lab assignments based on in-class coding grades. I show the applicability of using in-class coding to predict various final grade targets. Then, I determined the viability of using in-class coding as features compared to and combined with more traditional features, such as homework assignments, as well as the benefits to such an approach. While testing this I compared the traditional prediction of grades with the prediction of general performance, an easier to utilize feature for determining which students need help. I compared predicting curved grades versus uncurved grades to determine the effects of curving on student performance prediction. Finally, I did a deep dive into the features that could be generated from in-class coding to determine what features were most important to an accurate prediction, so that models could be run more quickly by using less total features.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest.

Notes

Open access

Download

Share

COinS