Unified fire sensing concept and time-series regression machine learning of environment-based features for quantifiable "no-fire" models

Abstract

This paper presents an in-depth exploration into fire sensing methodologies, with a specific focus on the development of a robust "No-Fire Detection Model." Various machine learning models were employed to construct a reliable predictive framework. The study emphasizes the identification and quantification of stable environmental conditions indicative of the absence of fire incidents. By leveraging time-series regression techniques and environment-based features, the proposed "Unified Fire Sensing Concept" effectively delineates boundaries for recognizing deviations in environmental parameters. This research seeks to advance fire detection systems by providing a comprehensive understanding of "No-Fire" modeling, offering insights into adaptable methodologies for enhanced safety and reduced false alarms.