Electricity Demand Forecasting For Residential Community: A Comparative Study of RF, KNN and DT Models

Abstract

Forecasting energy demand is crucial for residential communities in institutional buildings to optimize energy use and reduce costs. This study aims to develop a machine learning-based model for forecasting the energy demand of a staff quarter residential community in a campus of an institutional building in Kota, India. K-Nearest Neighbors, Decision Tree and Random Forest are selected for comparison because of their interpretability, computational efficiency and suitability for handling non-linear relationships in electrical demand forecasting. The model's performance has been evaluated using MAE and R2 performance metrics. The results show RF forecast outperforms with the lowest MAE of 0.3 compared to KNN (0.4) and DT (0.42) for optimizing energy management strategies within the site. The actual vs predicted load plots, error trend plots and error distribution plots help to understand model behaviour by means of dynamic prediction topologies. The integration of weather data analysis showcases seasonal patterns that affect energy demand. This work contributes to the growing field of smart grid optimization and sustainable energy management by offering a data-driven approach to forecasting energy use on-site for institutional residential settings which is a unique context as most existing studies focus on commercial or urban residential areas. A comparison with existing forecasting techniques shows that the compared model outperforms existing models in terms of accuracy thus providing a solution associated with peak load forecasting and opening the door to better smart grid performance.