LSTM and SHAP for transparent and effective IoB systems in IoT environments : household power consumption

Abstract

The integration of Long Short-Term Memory (LSTM) model and SHapley Additive exPlanations (SHAP) within the Internet of Things (IoT) lays the groundwork for transparent and effective Internet of Behaviour (IoB) systems, particularly in managing household power consumption. This thesis explores the development and validation of an LSTM-based forecasting model, employing SHAP for enhanced interpretability and user engagement within IoB frameworks. By leveraging minute-level household power consumption data, the study reveals the intricate dynamics of energy usage and provides a methodology for integrating explainable AI in household energy management systems. This integration helps the users to encourage the adoption of energy-efficient behaviours. The research findings indicate that LSTM, when coupled with SHAP, significantly improves the transparency and utility of IoB systems, offering actionable insights for energy efficiency and user behaviour modification. This thesis contributes to the discourse on sustainable energy practices, underlining the critical role of user-centric, interpretable AI in fostering energy-conscious behaviours and efficient resource utilization.