Skip to content
The UK's only independent think tank devoted to higher education.

Could AI help universities spot student success or failure earlier?

  • 4 September 2019
  • By Fred Singer

This blog has been contributed by Fred Singer, CEO of Echo360.

As the new academic year begins, universities across the country will be putting strategies in place to help reduce the 7.6% of full-time, first degree entrants at English higher education institutions who are no longer studying their courses the following year. Disadvantaged students are 3 percentage points more likely to drop out than their more advantaged peers.

But a new piece of research has suggested that AI could play a role in helping universities to predict students’ success or failure – with 72% accuracy – within the first 15 days of starting their course.

So, could AI help identify which students need what help, so the right support can be offered, at the right time, to ensure they make progress?

Uncovering barriers

Imagine Jess, a first-year Engineering student who has been relatively engaged in what she is learning from day one. She contributes in class and seems to enjoy her lectures. Jess, on the face of it, is the type of student that gives lecturers little cause for concern.

But according to the research, conducted by Dr Perry Samson, Professor of Climate and Space Sciences and Engineering at the University of Michigan, Jess’s behavior, or the way she learns in the first few days and weeks of joining her course, could provide an early warning that she might already be struggling. Jess’s lecturer, who is focused on explaining the complexities of mathematical theory to a class of over a hundred students, may not be in any position to spot the signs.

The study was conducted across seven semesters between 2015 and 2018 at the University of Michigan in the United States. The findings offer much food for thought for higher education institutions in the UK wanting to help their students learn more effectively.

The research looked at data from a number of the university’s systems to ascertain the learning patterns and grade point average of a cohort of 1,283 students enrolled in entry-level STEM courses.

It revealed that behavioral signals in the first couple of weeks, such as the number of correct responses to in-class questions, the number of notes taken, and the number of times students viewed slides and video recordings on the university’s learning platform could be used weekly to accurately monitor and predict student success.

The study is supported by Dr Louise Robson’s recent research in the Department of Biomedical Science at the University of Sheffield, which monitored student use of video lecture captures and participation in active learning activities and formative quizzes. Students who took part in formative quizzes were more likely to do well in their assessments. Those who used lecture captures and participated in active learning sessions were again more likely to do well.

Interestingly, students who had not done well were shocked when they saw how their use of the digital resources compared to their peers. The findings suggest that struggling students do not necessarily recognise that their learning approaches are inappropriate. This raises the possibility that by encouraging students to change their strategies, universities can better help them succeed.

These findings suggest that as the technology continues to advance, AI could offer universities a reliable method of spotting more students who need additional help much sooner than might be possible currently.

A different perspective

The vast majority of universities today use information such as assessment grades to monitor students’ achievement at regular points as the course progresses. Grades can provide an accurate indication of whether a student has understood the topics covered and help lecturers to spot students who could benefit from some study tips or extra support.

What the research shows is that if universities also monitored specific student behaviours, they could predict academic success – or prevent potential failure – even before the first exam.

This could help them to take a much more tailored approach to raising achievement. Giving students more time in class to cover complex topics in the early days, more opportunities to work collaboratively on problem-solving tasks and providing revision tips that fit with the way they learn could help students like Jess get maximum benefit from their studies.

The tools

What the research shows is the potential for information on learning behaviours to be used – with the relevant permissions – to support students more effectively through their course. The findings also provide the scope to explore how aspects such as gender differences might impact on achievement and what effect learning behaviours can have on closing attainment gaps.

Giving instructors insight like this at their fingertips makes it easier for them to intervene early and provide the most appropriate help to those who need it, in line with their background, previous achievement and current circumstances.

Could this AI-enhanced approach to assessment in higher education help students like Jess – and others such as those with special needs or studying in a second language – to better understand what steps they need to take to study more effectively for success?

1 comment

  1. Thanks for this post Fred,

    We’ve been working in machine learning space now for over 6 years with a number of UK universities in the development of Learner Analytics using ‘behavioural’ data, we have proven consistently over that time that data created during the learning process, especially when blended across a number of different data points (including learning data) we can create a powerful means of identifying risk within a student body.

    We have been extremely successful at stratifying risk and mobilising different stakeholders (both staff and students), around disparate data that support a number of institutional agenda’s such as wellbeing, progression, widening access and the overall student experience.

    We have found that our ‘engagement analytics’ provides a means to codify what is unique for each institution, school/faculty/dept and even course which allows for agency with both staff and students to intervene early on negative trajectories.

    We have evidence that video replay/lecture capture technologies have important parts to play in student outcomes, especially during critical phases of learning and we’d welcome the opportunity to integrate your data points in our StREAM software to determine if it can uplift the predictability of our algorithm.

Leave a Reply

Your email address will not be published. Required fields are marked *