Control approaches for micro-grid (MG) systems are recently developed for efficient energy management in distributed systems. The aim is to increase the integration of renewable energy sources (RES) in buildings while keeping optimal operational conditions of storage devices. However, the variability and the unpredictable behavior of the power produced by RESs require the use of energy management systems and adaptive control strategies for their seamless integration within the traditional electric grid. In this paper, a model predictive control (MPC) strategy is developed, named MAPCAST, for measuring, analyzing, predicting, and forecasting actions in order to ensure efficient and optimal operation of MG systems. The control strategy is based on machine-learning algorithms to predict main parameter inputs, which are used for forecasting suitable actions. Its main objective is to manage the batteries' charge/discharge (C/D) currents, and consequently, the battery state of charge (SoC), taking into consideration the variable nature of RES generation and loads demand satisfaction. A real data-set was gathered from our actual MG system using an IoT/Big-Data platform, which was deployed to measure the different input control parameters. Simulation results are presented to show the utility of the proposed control strategy for efficient operation and optimal energy balance in MG systems.