Forest Fire Prediction Frameworks using Federated Learning and Internet of Things (IoT)

Abstract

Forest fire ignition prediction is of paramount importance in safeguarding communities. Machine learning is a promising tool to enhance this. There is, however, a scarcity of publicly accessible forest fire ignition datasets. Moreover, existing platforms, such as the Canadian Wildland Fire Information System, underutilize machine learning capabilities for forest fire ignition classification prediction. Addressing the lack of forest fire ignition datasets and the untapped potential of machine learning, we propose two novel frameworks. One generates forest fire ignition prediction datasets for diverse geographic regions. The other incorporates federated machine learning and Internet of Things technologies to enable efficient and accurate forest fire ignition classification prediction. We evaluate the effectiveness of datasets instantiated from our data framework using various machine learning models. Additionally, using a simulated federated machine learning system we demonstrate prediction accuracy comparable to a centralized system with superior spatial granularity in classifying forest fire ignitions.