A Trustable Federated Learning Framework for Rapid Fire Smoke Detection at the Edge in Smart Home Environments

With the rapid growth of the Internet of Things, sensors have become integral components of smart homes, enabling real-time monitoring and control of various aspects ranging from energy consumption to security. In this context, we cannot underestimate the importance of sensor-based data in ensuring the safety and well-being of occupants, particularly in scenarios involving early detection of fire outbreaks. We propose a novel federated learning (FL) Framework in this study to address the crucial issue of rapid fire smoke detection at the edge of smart home environments. The proposed framework employs three distinct FL algorithms, namely, federated averaging, federated adaptive moment estimation, and federated proximal, for global aggregation of machine learning predictions based on data from various IoT sensors. This framework allows for early prediction by utilizing the computational capabilities at the edge, thereby improving the responsiveness and efficiency of fire safety systems. Furthermore, to improve trust and transparency in the FL framework, explainable artificial intelligence techniques, such as local interpretable model-agnostic explanations (LIMEs) and Shapley additive explanations (SHAP), are integrated. We unveil pivotal features driving predictive outcomes through LIME and SHAP analyses, offering users valuable insights into model decision-making processes.