IIoT and Industry 4.0 have enabled, in turn, revolutionary approaches toward energy
management for commercial buildings by giving way to more intelligent data-driven
decision-making mechanisms. The work presented in this paper aims to discuss the role
of electronic sensors and embedded systems as an enabler for real-time forecasting
of energy consumption in IIoT-enabled infrastructures. The three most common machine
learning algorithms, XGBoost, LightGBM, and Random Forest, were compared in comprehensive
detail with all types of sensor data: electricity, chilled water, hot water, and steam.
Unlike previous research, which has usually focused on a single model or limited evaluation
criteria, the research compared multiple models on seven different performance metrics
such as RMSE, R2, MAE, SMAPE, MSLE, F1 Score, and Precision. The results have shown
that XGBoost has outperformed other models consistently in terms of accuracy, proving
its potential in improving energy efficiency and enabling predictive maintenance schemes
in smart buildings. These observations enhance overall understanding in terms of effective
incorporation of machine learning models into IIoT-centric energy management systems
and provide considerable guidance for further development in energy-efficient infrastructure.
