Accurate prediction of building energy consumption for improving efficiency and reducing simulation costs requires an appropriate representation of occupant presence. In this study, occupant presence patterns at an hourly scale were encoded as continuous variables, and the months of the year were encoded cyclically (using sine and cosine functions) and used as inputs to EnergyPlus simulations and machine learning models. This approach, in contrast to discrete definitions or fixed occupancy rates, enables the analysis of the differing effects of presence across hours and seasons and improves the model’s ability to generalize to new conditions. After designing an artificial neural network capable of accurately predicting the monthly average total HVAC electricity demand rate, SHAP analysis and partial dependence plots showed that midday and afternoon hours have the greatest impact on energy loads and further revealed that changing the level of occupancy at different hours of the day, across winter and summer, has distinct and markedly nonlinear effects on energy use. Finally, it was shown that shifting the start time of occupancy, without any physical changes to the building, can optimize energy consumption and provides an occupant-centric framework for intelligent building energy management. Finally, the results indicated that optimizing the timing of occupied windows throughout the year for a simple schedule can lead to reduction in the monthly average demand rate up to approximately 15%.