MASc Seminar: Predicting student performance using data from an auto-grading system

Candidate: Huanyi Chen

Title: Predicting student performance using data from an auto-grading system

Date: December 15, 2017

Time: 11:00am

Place: EIT 3142

Supervisor(s): Ward, Paul A.

Abstract:

In engineering education, predicting student performance is significant. Better balancing the resources put into each student instead of dividing the recourses equally is significant to institutions. As online auto-grading systems appear, information obtained from those systems can potentially enable us to create predictive models to predict student behavior and performances in a fairly early stage.

In this thesis, we conducted several experiments with decision tree and linear regression algorithms using the information from an auto-graded system Marmoset aiming to answer the questions: 1. If we put students into categories according to their performances and our pre-knowledge, can we predict their categories? How accurate is that? 2. Can we predict students' raw numerical midterm grades or final grades? If so, how early can we predict that? 3. Can we find any interesting relations between features generated from auto-grading system information, grades and student categories? We found out the best result come from applying the linear regression algorithm to the time interval between the student’s submission and the assignment deadline. We achieved 0.78 correlation coefficient in predicting the raw final grade and 69.76% accuracy in predicting student categories.