The availability of quality indicators in all areas has increased dramatically in recent years. At the same time, this growing volume of data is leading to difficulties in accessing and digesting this information, resulting in what is known as KPI overload - an overabundance of indicators that can even become an obstacle to achieving real improvement. This led the agency to begin exploring machine learning techniques, together with legislative changes that put the focus of assessment on the university centre rather than the degree. By analysing large amounts of data, AI algorithms can identify patterns and predict potential risks to degree quality, becoming a useful complement to university centre diagnostics.
Specifically, the study analysed more than 70 indicators related to access, academic progress, student satisfaction and employment, as well as more than 1,200 evaluations carried out between 2015 and 2021. Machine learning techniques were therefore used to establish a relationship between these indicators and the history of quality assessment scores, assessing their potential predictive power.
By analysing large amounts of data, AI algorithms can identify patterns and predict potential risks to degree quality, becoming a useful complement to university centre diagnostics
The analysis showed that, with some exceptions related to the teaching staff quality dimension, the correlation between the indicators and the assessment outcomes was generally weak. Despite this weak relationship, the machine learning model showed a high rate of specificity in predicting conditional accreditation outcomes: it was able to correctly predict the risk outcome (i.e. accreditation with conditions) of 99% of degree assessments and 100% of Master's degrees.
This exploration, despite its potentialities, is not without limitations. On the one hand, the history of evaluation results used to train the algorithms is rather limited, given the diversity of the reality between disciplines, modalities and ownership of centres. On the other hand, the need to keep the models up-to-date with regulatory or methodological changes (i.e. institutional accreditation) is an issue that cannot be ignored in the future.
It is important to point out that quality assessment can never become a purely automatic process carried out by machines, but has a strong qualitative component that cannot be dispensed with. However, the use of artificial intelligence in quality assessment is an interesting tool to help identify potential risks and complement quality assurance processes, both for universities and for quality assurance agencies. Therefore, despite its limitations, this research opens the door to a future in which technology can play a key role in the continuous improvement of higher education, and one that needs to be further explored.
.jpg)