<p>The Machine Learning (ML) algorithms utilized in forecasting energy use in smart homes have enhanced the sustainability and efficiency of energy utilization. Dynamic energy management can be feasible since ML algorithms will take hold of past trends to predict future demand. Thanks to the model's predictive power, energy resources can be utilized more efficiently, reducing waste and running expenses. The goal of the current work is to result in a rather precise forecast of energy consumption in smart homes using advanced ML techniques; a careful investigation is being carried out into the Random Forest (RF) and Voting Regression models. These base models were combined with sophisticated optimization methods such as Bonobo Optimizer (BO) and Dynamic Differential Annealed Optimization (DDAO) to enhance these forecasted results. New hybrid models resulted from this integration, which outdid the earlier ones, as evidenced by reduced root mean squar error (RMSE) values and higher R-squared metrics. Extensive training, validation, and testing were done to compare the different hybrid models for the best in forecasting smart home energy use. While the RFDD model turned out to be the best, with a lower RMSE of 0.172 during testing, the VODD model predicted at a moderate level of accuracy with a value of 0.194. The RFBO model, on one hand, did exhibit lesser efficacy with an RMSE of 0.223. These findings represent key new information for the smart home energy management research area and stress the importance of model tuning in improving predictive performance.</p> Graphical Abstract <p></p>