The increasing use of renewable energy sources has presented challenges for energy management in IoT-based systems. Efficient energy management is essential for enhancing overall system performance. Machine learning techniques, such as Load Forecasting (LF), are increasingly applied to predict and regulate energy consumption. However, many existing methods fail to adequately address prediction errors and accuracy in short-term intervals. To bridge this gap, a study aims to develop a Regression model and compare its performance with Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) algorithms. The proposed model is designed to forecast energy consumption for the upcoming day and week, utilizing historical data from the Moroccan buildings’ electricity consumption dataset (MORED) dataset. Experimental evaluations on Hourly Energy Prediction (HEP) and Weekly Energy Prediction (WEP) tasks show the superior performance of the proposed Regression model compared to recent deep learning models. The results include HEP evaluations with a rootmean-square error (RMSE) of 0.603, MSE of 0.364, MAE of 0.5091, MAPE of 13.60%, and an accuracy of 86.40%. For WEP evaluations, the RMSE was 0.071, MSE was 0.0052, MAE was 0.0657, MAPE was 1.60%, and the accuracy was 98.40%. The study recommends this model for real-time datasets in Africa.