The increasing demand for energy-efficient electric gadgets has motivated the development of advanced technologies that can adapt to varying operating conditions while minimizing losses. There is a crucial need for energy-efficient induction motor systems, as the applications of electric motors continue to expand rapidly across industries and daily life. In this paper a novel system architecture is presented featuring a semiactive variable inertia flywheel (VIF) capable of dynamically adjusting its moment of inertia by radial mass displacement to enhance energy efficiency and adaptability with electric motor systems. Experimental validation demonstrates that the integration of a magnetorheological (MR) fluid-based VIF significantly reduces average power consumption at both rated and subrated speeds. However, the used complex non-Newtonian behavior of MR fluids poses challenges to achieving accurate real-time prediction of power consumption using conventional methods. To address this, machine learning algorithms are employed to predict optimal semi-active VIF control parameters, enabling real-time dynamic adaptation under diverse operating conditions. The results show a strong agreement between the simulation and the experimental data, confirming the accuracy and reliability of the proposed model. The machine learning framework effectively captures the relationship between system parameters and energy efficiency, paving the way for intelligent, energy-aware electric drive systems in practical applications.