Detection of Electrical Fires in Residential Buildings Using a Gradient Boosting Machine Algorithm

Abstract

The paper discusses a technique for detecting electrical fires in residential buildings using the Gradient Boosting
Machine (GBM) algorithm. The features of the algorithm process data related to current, voltage, and total harmonic
distortion (THD) from electrical systems, considering resistive loads, such as heating appliances, and inductive loads,
like refrigerators and washing machines. The technique underscores the relationship between electrical characteristics
and fire risks, demonstrating that the gradient boosting machine can accurately predict fire hazards under various fault
conditions, including arc faults, overvoltage, and contact opening. Results from MATLAB simulations confirm the algorithm's efficacy and high accuracy rates for heating systems and induction motors across different fault types that
could lead to electrical fires in buildings. These results highlight the significance of effective feature selection in
enhancing the algorithm's performance while addressing some imprecision, particularly regarding the two different
load types. Ultimately, the Gradient Boosting Machine represents a promising approach to improving the safety of
electrical systems and supporting fire detection strategies.