Machine Learning Approaches for Residential Electrical Fire Risk Assessment: A Review

Abstract

The risk of electrical fire remains high in residential environments because of ageing installations, rising electrical load demand, and the prevalence of 'informal wiring'. Conventional fire detection systems are reactive; they do not work before fires start. With the recent developments in machine learning and the Internet of Things (IoT) technologies, predictive models are now able to detect pre-ignition conditions. This paper aims to review machine learning algorithms that have been used for the assessment of electrical fire hazards in residential settings. Algorithms reviewed include Random Forest, Support Vector Machines (SVM), Decision Trees, Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM) networks, and Artificial Neural Networks (ANN). The algorithms assessed include the ability to predict well, be easy to understand, to be fast, and to be appropriate for data that are not evenly distributed. Results reveal that the ensemble methods such as Random Forest and XGBoost have produced good prediction performance. Random Forest is also a promising solution because of its interpretability and robustness when working in safety-critical environments. Evidence is still varied and weak, though, to conclude that any specific model is superior. It is a narrative review, and some important research gaps are highlighted, such as the lack of data specific to the residence and short-time modelling. The future research directions are focused on explainable AI, transfer learning and domain-specific validation.