This research paper focuses on using a data-driven machine learning model to predict hourly space heating demand in residential houses for predictive and supervisory control purposes. It discusses current energy industry practices in optimizing energy usage and costs in buildings using Machine Learning (ML) techniques, highlighting different model types like white-, gray-, and black-box models. White-box models, based on physics, are challenging due to complex system dynamics and numerous parameters, leading to higher energy consumption, costs, GHG emissions, and reduced comfort. Model Predictive Control (MPC) and AI models are currently being used as alternatives to white-box models, offering faster steady-state attainment and lower offsets compared to PID controllers. A case study in Toronto, ON, involved forecasting natural gas furnace run time using a dataset from a Wi-Fi-connected smart thermostat and climatic data from NASA and ECCC. Various deep-learning models (RNN, LSTM, GRU) and regression models (SVM, GBR, RFR, MLPR, Ensemble) were developed and fine-tuned, achieving a maximum R-squared value of 0.91, MAE of 0.098, and MSE of 0.2 on the test dataset.