An Ensemble-Based Regressor Framework for Energy Usage Prediction in Sustainable Building Management

Efficient energy consumption is a critical concern in sustainable building management, driven by increasing energy demands and environmental considerations. This paper focuses on developing a predictive model to analyze and optimize energy usage in buildings using advanced machine learning techniques. The research integrates diverse factors such as building type, operational status, maintenance priority, air quality, and smart device count to provide actionable insights into energy consumption patterns. A structured approach to feature engineering, including encoding categorical variables, handling temporal data, and removing irrelevant features, ensures robust model performance. This study focuses on the prediction of energy consumption in buildings using a voting regressor model, which integrates Random Forest Regressor, catboost regressor, and Multi Layer Perceptron regressor. The proposed voting regressor combines the strengths of the base models, leveraging their complementary capabilities to enhance prediction accuracy. Experimental results indicate that the proposed regressor model consistently outperforms individual models and other ensemble approaches in terms of prediction accuracy and generalization. The integration of tree-based and kernel-based regression methods enables robust performance across diverse building characteristics.