Universities face growing pressure to optimise space use amid tightening budgets, declining student numbers, and rising energy costs, yet educational spaces remain chronically underused. This study examined whether course and spatial characteristics can predict two space performance metrics, space utilisation and educational productivity (ECTS credits generated per scheduled hour), and whether historical patterns or these characteristics carry more predictive weight. 

Using four years of administrative data from the UvA, courses and educational spaces were profiled to train several machine learning models. Historical patterns, when available, proved highly predictive; for utilisation, so much so that they outperformed the trained models, whereas for productivity the models still added value beyond history. Critically, the interaction between courses and rooms mattered most when no historical data was available, showing that it does matter for specific factors which courses are given in what rooms. These findings give insights into why it matters where you holda course and builtsinstitutions data-driven foresight for space and curriculum planning.