Smart homes incorporate numerous devices that automate processes and simplify lives, assisting in managing temperature, lighting and security access. The capacity of smart homes to learn and adapt to human owners’ routines is powered by the Internet of Things (IoT) network. Due to the multitude of devices and the necessity for efficiency and cost savings, scrutinizing power consumption for each device is crucial. Furthermore, higher energy consumption contributes to a larger carbon footprint, greater climate risk and higher supply demands, necessitating monitoring of energy usage. We employ sophisticated machine learning algorithms for an exhaustive study of energy consumption in smart homes, introducing a novel sailfish-optimized random forest algorithm (SORFA) to predict energy usage. SORFA optimizes the hyperparameters of the random forest algorithm to enhance its predictive accuracy and efficiency. Our proposed method, when compared with baseline methods, demonstrates superior performance in estimating energy consumption by reducing computational time and improving prediction accuracy. These improvements suggest significant potential for energy savings and cost reduction in smart home environments.