Owen Corrigan - Transfer Talk - 15th May 2015

Video Category: 
Transfer Talk
OwenCorrigan

Title: Predicting Learner Achievement using Resource Access Logs

Supervisor: Prof. Alan Smeaton

Abstract:

This thesis aims to use data generated by students as they are using online educational systems to try to answer the question: in what ways can we use student  data to improve their
learning experience.  This question came about as a result of a system I built to predict students' exam performance based on features extracted from their usage of the Moodle/Loop
system, features such as time spent online, use of resources, and typical days and times when accessing the system. We found that it was possible to build a predictor which accurately
classified whether a student would pass or fail a module. This classifier would get more accurate over time as the students interacted with Moodle/Loop more, and as we got closer to the
exams. As a form of feedback, students and their Lecturers received weekly emails that warned them if they needed to study more, if that was appropriate.

There are many open questions raised by this preliminary project which I intend to investigate as part of my PhD thesis. These include

How can we validate the predictors when there is a constant stream of incoming usage data ?
How can we integrate other sources of data besides Moodle/Loop access logs into this system ?
How can we validate if the predictive analytics feedback to students worked, i.e. did the students grades improve, and if so, and how much of this was due to the feedback ?
What effect does the students adapting their behaviour to the system have ?
Why are some courses not good candidates for predicting student performance, and how can we improve predictions for them ?
Can we target the students in more specific ways in terms of the content they should view. Can we tell them things like ``I recommend you study chapter 2 more'' ?
What is the behavioural effect on students of being rated against their peers and how can this feedback be adapted to provide useful and actionable information to lecturers ?
What are the ethical implications of this kind of analysis ?

This seems like a lot of questions to be asking at this stage of a PhD and certainly I will not be able to address all of them. They all fall under the one hypothesis about using student data
to improve the learning experience but the specific ones I will address will depend on the results from the system I built in DCU for Semester 1 and 2 of this year. The analysis of the
impact of that will determine which questions I will focus on.

Finally, my thesis is a big data analytics application at the intersection of three sub fields which have been growing rapidly in recent years. The first in Learning Analytics, which studies
data from learners and their contexts. The second is Educational Data Mining, which explores how data generated from student behaviour can be used. The third is gamification, which is
the use of game thinking in non-game contexts to engage users in solving problems.