Trustworthy Assessment of University Competitiveness Using a Neural Network Model

Universities compete for funding, and their positions depend on the results of national assessments and rankings, which are expensive to produce and based on difficult-to-predict expert opinions. Assessment results have a significant impact on a university’s reputation, funding levels, attractiveness to faculty and staff, and success in recruiting top-tier students. Expert assessments and forecasts are widely used, but additional support from trusted AI tools is desirable. Several attempts have been made to use various machine learning methods, but confidence in such solutions is limited due to perceived difficulties in clearly and reliably justifying the resulting predictions. This research aims to present a proposal for using neural network models, accompanied by explanations of their predictions, to support trustworthy and sustainable assessment of university competitiveness. This methodological contribution enhances the transparency and interpretability of the assessment process and is further supported by empirical studies based on data from selected universities. A Fully Connected Neural Network (FCNN) is used for the calculations, and the local interpretable model-agnostic explanations (LIME) method is applied to explain the prediction results. The results confirm the usefulness of the proposed model and provide a solid foundation for improving evaluation systems and building trust in AI applications for assessing universities’ competitive position and the benefits of scientific research for society.