Design of a Hybrid Model Based on Statistical and Machine Learning Techniques for Effective Forecasting of Smart Home Energy Consumption

Abstract

Smart home technology is widely used across all cities to provide high security, comfort, and energy efficiency. This leads to an increase in energy consumption due to the growth in technology, population, and economy. This creates a necessity for forecasting energy consumption to fulfill the unrelenting demands of electricity consumers. In this view, researchers have implemented many statistical methods (namely autoregressive integrated moving average (ARIMA) and seasonal ARIMA (SARIMA)), machine learning models (namely decision trees, random forest, and gradient boosting), and deep learning models (namely artificial neural networks (ANNs), and long short-term memory (LSTM)) for forecasting. However, the implementation of these individual methods often produces inaccurate forecasting results. As a solution, this paper proposes a hybrid model that combines the methods of ARIMA and Extreme Gradient Boosting (XGBoost). These are considered for hybridization due to their simplicity in implementation when compared to other complex models. Usually, ARIMA captures only trends and seasonality, while XGBoost handles only complex and non-linear patterns. Thus, the proposed hybrid model solves these issues in individual models, producing superior results. For validating the effectiveness of the proposed hybrid model, various performance metrics, namely RMSE, MAPE, and R-squared, are computed and compared with the values obtained with the ARIMA and XGBoost when considered individually. These computations gave RMSE of 40.3% (ARIMA), 36% (XGBoost), 33.2% (proposed), MAPE of 48.84% (ARIMA), 53% (XGBoost), 37.26% (proposed), and Rsquared of 0.3 (ARIMA), 0.2 (XGBoost), 0.4 (proposed), thus the superiority of the proposed hybrid model is verified. This research work is implemented using a "smart home dataset" from Kaggle.