Buildings consume about 40% of the energy produced globally, with a significant portion being used by indoor air conditioning systems. Given the threat of rising global temperatures, the use of Heating, Ventilation, and Air Conditioning (HVAC) systems is projected to increase, further reducing overall building’s energy efficiency. Retrofitting building energy systems and envelopes has the potential to mitigate anticipated future energy use and discomfort caused by extreme weather events such as heat and cold waves. In this study, we integrated an ML-based window state prediction meta-model into the large-scale models published in Model America V1 (MAv1) by Oak Ridge National Laboratory (ORNL). Homes in all counties of New York State (NYS) were selected using a random sampling method, constrained to buildings with characteristics matching those used to train the meta-model. The meta-model was developed using a pre-trained series of window state ML models from a residential dormitory in Upstate New York. The original IDFs were modified with Python to include the new objects required for simulation. These modifications enabled the generation of variables and actuator handles needed for the ML meta-model to function. Additionally, a natural ventilation (NV) model was incorporated into all IDFs, as the original files did not include this feature. Future weather files for the year 2094 were obtained from ORNL and assigned to each randomly sampled IDF using the haversine function based on geographic proximity. Simulations were conducted using the EnergyPlus Python API to analyse changes in NV usage and the corresponding impact on energy consumption in 2094 compared to present-day weather conditions. Clustering analysis revealed that NV usage in the year 2094 may vary between −79.46% and 68.97% compared to the present. Similarly, cluster analysis for the corresponding energy use indicates that changes ranging from −10.22% to 3.56% compared to the present are possible.