The study investigates two predictive models, XGBoost (a machine learning method) and Bi-LSTM (a deep learning approach), for the task of predicting hourly electrical consumption. The study uses Vermont, USA, data covering the period 2015-2025, with the two models trained on electrical consumption and weather data. Their performance was assessed using error metrics such as MAPE, RMSE, MAE, and R². The results show that Bi-LSTM comes out on top of XGBoost, with the results being more accurate, having lower error rates. Although XGBoost does okay for very short-term predictions of up to 6 hours, Bi-LSTM outperforms for short-term and longer-term predictions of up to 24 hours. The outcome implies that deep learning models, such as Bi-LSTM, are more effective in capturing complex patterns of electrical load data in time. In short, the study shows that the two models are both able to deliver; however, Bi-LSTM has a clear edge for sustainable electrical demand forecasting, thus enabling better energy planning and sustainable power production.