Machine Learning Regression Approach for Estimating Energy Consumption of Appliances in Smart Home

• Main Authors: Husin N.S.I.M., Mostafa S.A., Jaber M.M., Gunasekaran S.S., Al-Shakarchi A.H., Abdulsattar N.F.
• Other Authors: 58581629000
• Format: Conference Paper
• Published: Institute of Electrical and Electronics Engineers Inc. 2024
• Subjects: Appliances Energy Estimation; Energy Management; Machine Learning; Time Series Dataset; Automation; Data mining; Decision trees; Energy conservation; Energy management; Forecasting; Learning algorithms; Learning systems; Machine learning; Regression analysis; Time series; Appliance energy estimation; Energy estimation; Energy-consumption; Machine learning algorithms; Machine-learning; Smart homes; Three models; Time series dataset; Times series; Training and testing; Energy utilization

This paper attempts to use machine learning algorithms to estimate the energy consumption of appliances in a smart home environment. This work aims to promote awareness among smart home systems and users about their appliances' energy consumption and guide them toward energy-saving practices. To achieve this, three machine learning algorithms, namely Decision Forest (DF), Boosted Decision Tree (BDT), and Linear Regression (LR), were chosen for regression tasks to estimate the energy consumption of several appliances accurately. The time-series datasets, namely appliance energy prediction datasets, are used for training and testing the algorithms. The Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology, which comprises six processing phases, was employed in this work. The test is performed by utilizing 10-fold cross-validation. The results obtained assess the models' performance in predicting the appliances' energy consumption. The experimental results indicate that the three models exhibit varying degrees of accuracy in predicting energy consumption, as measured by their respective R-squared values. Among the three models, the random forest model exhibited superior performance by achieving the highest R2 values of 0.62 and 0.54 during the training and testing phases, respectively. � 2023 IEEE.