Energy Consumption Prediction in Smart Homes Using QIO-Enhanced Regression Models

Abstract

Accurately forecasting household energy consumption remains a challenge due to the variability introduced by user behavior, appliance diversity, and environmental conditions. Smart homes have also become a key solution to the expanding global energy demands in the residential arena through the introduction of new strategies to maximize and control the use of energy. Smart systems comprise sophisticated sensors and networked appliances to provide accurate, affordable, as well as ecofriendly management of energy. Yet, forecasting the use of energy in such homes is challenging with the differences in end-user behavior, weather, and appliance efficiencies. In this work, the use of machine learning (ML) algorithms Extra Tree Regression (ETR), Naive Bayes Regression (NBR), and Elastic Net Regression (ENR) to predict the use of energy in a home is presented. These are also optimized with Quadratic Interpolation Optimization (QIO) to adjust their hyper parameters. Experiments were performed using the Kaggle Smart Home Energy Usage dataset, which provides comprehensive synthetic data across various features of occupancy, appliance usage, temperature, and timestamped consumption data. The developed hybrid schemes were tested using the criteria of R² and RMSE in the training, validation, and testing stages. Out of the varied modeled algorithms, the ETQI (ETR + QIO) model realized the highest accuracy of R² = 0.985 and RMSE = 0.245 while outperforming NBQI (NBR + QIO) with R² = 0.974 and RMSE = 0.233 and ENQI (ENR + QIO) with R² = 0.952 and RMSE = 0.319. These conclusions reflect the ability of optimized ML algorithms to drive more effective energy efficiency strategies in the smart home scenario.