Towards Lightweight Fire Detection at the Extreme Edge based on Decision Trees

The prevalence of fires in modern urban landscapes continues to be reported as a significant challenge due to their potential negative consequences. This enduring threat heightened by climate changes necessitates the development of innovative approaches for early detection, incorporating a variety of sensor-based technologies across diverse scopes. Although effective, traditional detection methods based on Unmanned Aerial Vehicles (UAVs) or camera sensors often encounter cost, scalability, and deployment limitations, particularly for vast areas to be covered. On the other hand, scalar sensors for threshold-based fire detection are limited to the context and require proper calibration for different scenarios. This paper addresses these limitations by introducing a novel approach that leverages a sensor-based decision tree model for fire and smoke detection using micro Internet of Things (IoT) devices. The core of this approach lies in developing an efficient, lightweight model suited for extreme edge devices, where performance and energy efficiency are critical. We validated our model using a public dataset, demonstrating its potential effectiveness in real-world scenarios. By providing a sensor-based solution that is both scalable and cost-effective, the proposed solution emerges as a promising tool for augmenting fire safety measures in urban and wildland areas.