Machine learning optimization for energy consumption prediction

Global energy use has increased significantly over the last few decades, with residential structures accounting for a sizable amount of this usage. Therefore, creating trustworthy instruments for assessing and predicting energy use has become crucial in the global endeavor to improve sustainability, ultimately leading to more effective energy management strategies and a reduction in greenhouse gas emissions. Machine learning (ML) techniques have demonstrated high accuracy in energy usage prediction tasks. Using the publicly accessible KAG energy dataset, we assess and contrast ten ML and deep learning (DL) models to forecast energy usage in smart buildings, including Extra Trees Regression (ExtraTr), Long Short-Term Memory (LSTM), Multi-layer Perceptron (MLP), XG Boost Regressor, Gradient Boosting (GBoost), Convolutional Neural Network (CNN), Random Forest Regressor (RF), Elastic Net (ElNet) Regressor, Polynomial Regressor, and Support Vector Regresssor (SVR). The obtained results were compared with those of ARIMA model. According to our experimental findings, LSTM outperforms other models in capturing temporal dependencies in energy consumption data, demonstrating its superior ability to capture longterm patterns and fluctuations. This suggests that the recurrent nature of LSTMs, which allows them to retain information about past energy usage, is crucial for accurate forecasting in smart buildings.