A modern workflow for energy consumption prediction: Comparative analysis of machine learning techniques

In smart-campus environments, precise forecasting of electricity demand is crucial for enhancing energy planning and operational efficiency. In this research, a machine learning framework is developed to forecast short-term electricity consumption by integrating high-resolution smart-meter data with meteorological variables. The workflow encompasses data preprocessing, outlier management, and feature engineering. This includes creating an operational ON–OFF indicator that reflects laboratory activity and short-term consumption variations. Several regression models; Bootstrap-based 95% confidence intervals were used to quantify uncertainty, and residual analyzes offered additional insights into model reliability. The proposed framework is efficient in terms of computation and scalable, providing a strong and transferable approach to facilitate data-driven energy management in academic institutions.