Exploiting scalable machine-learning distributed frameworks to forecast power consumption of buildings