This thesis comprehensively investigates forecasting building occupancy levels using the Koopman Operator in the context of the university canteen at the Technical University of Munich. The project attempts to improve occupancy prediction algorithms by applying Extended Dynamic Mode Decomposition and comparing them to established and innovative forecasting methodologies. The study uses a manually created dataset, based on Wi-Fi sensor measurements, which also uses meteorological variables and time embedding techniques, to estimate occupancy trends. The findings show that the Koopman-based methodology is successful at capturing dynamic occupancy changes, showing promising prediction accuracy compared to the other approaches. This study not only investigates time-series forecasting approaches but also delivers practical insights for campus facilities management, demonstrating the value of data-driven decision-making in enhancing canteen operations. Thus this thesis contributes to the fields of machine learning and campus activities. Future research directions include investigating integrating other data sources and developing the prediction model for more general applications.