How AI is helping disadvantaged UK young people get into top universities

The use of AI (Artificial Intelligence) has helped boost the chances of at least 150 young UK people from disadvantaged backgrounds getting into top British universities, since 2022. 

By using accurate data and a tweaked machine learning model, UK University access charity and tutoring organization, The Brilliant Club, is identifying individuals who are at danger of failing its pre-University support program.

A key metric, the ‘support score,’ has been used since Spring 2022 to flag up teenage scholars who might not otherwise have made it through the high-level degree entry course the Club offers.

Doing so ensures more students can complete its ‘final assignment’ - and so those who otherwise probably wouldn’t have secured slots in top tier universities get their life chances significantly improved.

This really matters, as in 2020 12% of the 8,000 participants that took the course did not submit that final assignment, meaning they did not complete.

Boosting applicant knowledge and confidence levels

Dr Lauren Bellaera, the Club’s Chief Impact and Strategy Officer, says:

Our mission is to support less-advantaged students to access university and then to succeed when they're down there. All our charitable endeavors are focused on that, and any help we can get doing that really helps.

The context here is what The Brilliant Club sees as addressing the challenge British youngsters from disadvantaged backgrounds face in getting into higher-prestige Higher Education institutions.

This centers on the Russell Group of research-oriented universities, which includes Oxford, Cambridge, and Kings College London.

Its data suggests just 2% of such applicants gain places in Russell Group universities, compared to 28% of the most advantaged.

It’s not so much a lack of innate academic ability holding them back, The Brilliant Club believes.

Instead, says Bellaera, it’s often down to too little information about the whole concept of university, how to apply, and their families and communities lacking familiarity with this world, too.

To try and boost their chances, Bellaera and her team ask the UK PhD community to pitch in and boost applicant knowledge and confidence levels - a strategy that independent assessment suggests may give a candidate a 60% better shot at acceptance.

This happens via PhD holders or researchers taking work as tutors in identified UK schools and delivering university-level classes to a selected group.

This in-class work is supplemented by two trips to a competitive possible destination University, one at the start of the application and one on completion of that key deliverable, the student final assignment.

She says:

A final assignment is an academic, challenging piece of written work and we have consistent data showing that if students complete it, they are statistically significantly more likely to progress to a competitive university.

The overall framework a final assignment needs to be written in is The Brilliant Club’s largest single university access initiative, The Scholars Program.

This now supports 16,000 students across the UK each year, and as a result, a dataset of previous applicants had been building up.

However, that wasn’t really being usefully ‘mined’ for insights, says Bellaera.

An obvious gap was a reliable way to predict who would struggle - or not - with that critical last step of submitting their final Scholars Program assignment.

She says:

Essentially the business problem here was trying to increase the completion rate of the final assignment so we could see who those students seem to be and then what support mechanisms we can put into place and therefore have the best opportunity to succeed academically and to progress into higher education.

Three years’ worth of Scholars Program pupil records

Bellaera stresses that The Club was already confident about data; the issue was capability and time to do more with it.

She says:

We have a lot of data, are very good with data. What we lacked was a way of using big data, AI and predictive modelling on it.

To try and get some answers, Bellaera decided to share three years’ worth of Scholars Program pupil records with the AI company it eventually decided to work with, Dataiku.

That company had offered to allow the non-profit use of its AI-for-good ‘Ikig.AI’ initiative, which grants organizations access to its core Everyday AI platform.

That dataset - some 70,000 student profiles and performance results - included basic demographics, engagement data, school, and information on each student from Brilliant Club tutors.

The Dataiku team then ran its predictive modelling software over this data to understand how to better support strivers applying for Brilliant Club support.

It also carried out a technology transfer where it helped the Club extend its in-house data analytical capabilities.

In turn, this is being helped by ongoing training and the pro bono assistance of a Dataiku data scientist.

The result, she says, has been a strengthening of the Club’s support system via new guidance and training for the 400 PhD-level tutors who deliver The Scholars Program.

But it also delivered a ‘high, medium, or low’ confidential ‘risk factor’ for each pupil, she says.

A new set of risk ratings 

Technically, this risk factor is built from application of a machine learning technique used in regression and classification tasks called gradient boosting.

The vendor and Club agree this technique delivered results based on past experience that was 81% accurate in predicting what factors had made previous Scholars Program candidates fail to complete.

But from her point of view as an educationalist, what really matters is that a new set of risk ratings and ensuing interventions for medium- and high-risk students were implemented for The Scholars Program.

This is delivered as a visual data tool that can easily be used by non-data scientist managers at the Club, Ballaera says.

They run the tool regularly to see what pupils may be struggling, and immediately suggest aid and interventions for those who are red flagged, she adds.

In terms of next steps, Ballaera stresses the building of the risk factor tool was for a very closely defined problem within just one of the Club’s programs.

Now that issue has been so successfully resolved, as the non-profit runs multiple university access and academic support and success programs, the principles and collaboration of this project can now easily be applied to help other parts of the Club’s mission.