Intelligent energy management in IoT-enabled smart homes: Anomaly detection and consumption prediction for energy-efficient usage

Laviza Falak Naz; Rohail Qamar; Raheela Asif; Saman Hina; Muhammad Imran; Saad Ahmed

doi:10.22581/muet1982.3291

Laviza Falak Naz Department of Software Engineering, NED University of Engineering & Technology http://orcid.org/0000-0003-3740-209X
Rohail Qamar Department of Computer Science & Information Technology, NED University of Engineering & Technology http://orcid.org/0000-0001-8697-6706
Raheela Asif Department of Software Engineering, NED University of Engineering & Technology http://orcid.org/0000-0001-5155-8205
Saman Hina Department of Computer Science & Information Technology, NED University of Engineering & Technology http://orcid.org/0000-0002-7649-4372
Muhammad Imran Department of Computer Science & Information Technology, NED University of Engineering & Technology http://orcid.org/0009-0004-9029-6526
Saad Ahmed Department of Computer Science, IQRA University http://orcid.org/0000-0001-6121-8124

DOI: https://doi.org/10.22581/muet1982.3291

Abstract

The increasing Internet of Things (IoT) device integration in smart home environments has increased the options available for intelligent energy management. In the context of smart homes, this paper provides a detailed analysis on the use of IoT data for energy consumption trend prediction and anomaly detection. We propose a novel approach that combines the advantages of the Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) models for accurate consumption forecasting. Real-world data from a smart home setting is utilised to evaluate the proposed models. Results will therefore show that our approach performs best in optimally utilizing resources, minimizing waste, and improving energy consumption. The current study contributes to the development of energy-efficient smart houses through providing a reliable method for consumption forecasting and anomaly detection. Results indicate that the LSTM model outperformed ARIMA in prediction accuracy, achieving a lower Mean Absolute Error (MAE) of 0.110 compared to ARIMA's 0.176. Furthermore, the LSTM model demonstrated superior performance in anomaly detection, with higher precision and recall scores.