This study evaluates data-driven models for forecasting one-day-ahead hourly electricity consumption in buildings to improve the operation of building energy management systems (BEMS), with the aim of improving cost efficiency, grid stability and reducing CO2 emissions. Five models were evaluated for forecasting energy use in a modern, well-insulated university building: Multiple Linear Regression (MLR), Holt-Winters Exponential Smoothing, Prophet, eXtreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM). The models were tested using energy consumption data from the Delft Building of The Hague University of Applied Sciences, including occupancy, operating schedules and weather conditions. The results show that Prophet and XGBoost outperformed the other models on weekdays on 4 metrics. Of these, Prophet gave the best performance on R2 score (0.92) and MAE (17.78 kWh), while XGBoost performed the best on the MAPE (14.12%) and MSE (794.36). These results for our data implied that Prophet had a strong ability to capture the underlying trend and seasonality while XGBoost better captured the the relative errors in the estimation. MLR performed similarly to XGBoost, highlighting the effectiveness of key predictors. The LSTM model showed weaker performance due to limited data and hyperparameter constraints, and Holt-Winters showed higher errors during low consumption periods. Predicting non-working days remained a challenge for all models. This study advances data-driven predictive modelling of building energy use to support efficient BEMS strategies. It is one of the few studies to use the Prophet model to forecast energy consumption.