Skip to main content

Research Repository

Advanced Search

Sustainable Collaboration: Federated Learning for Environmentally Conscious Forest Fire Classification in Green Internet of Things (IoT)

Siddique, Ali Akbar; Alasbali, Nada; Driss, Maha; Boulila, Wadii; Alshehri, Mohammed S.; Ahmad, Jawad


Ali Akbar Siddique

Nada Alasbali

Maha Driss

Wadii Boulila

Mohammed S. Alshehri


Forests are an invaluable natural resource, playing a crucial role in the regulation of both local and global climate patterns. Additionally, they offer a plethora of benefits such as medicinal plants, food, and non-timber forest products. However, with the growing global population, the demand for forest resources has escalated, leading to a decline in their abundance. The reduction in forest density has detrimental impacts on global temperatures and raises the likelihood of forest fires. To address these challenges, this paper introduces a Federated Learning framework empowered by the Internet of Things (IoT). The proposed framework integrates with an Intelligent system, leveraging mounted cameras strategically positioned in highly vulnerable areas susceptible to forest fires. This integration enables the timely detection and monitoring of forest fire occurrences and plays its part in avoiding major catastrophes. The proposed framework incorporates the Federated Stochastic Gradient Descent (FedSGD) technique to aggregate the global model in the cloud. The dataset employed in this study comprises two classes: fire and non-fire images. This dataset is distributed among five nodes, allowing each node to independently train the model on their respective devices. Following the local training, the learned parameters are shared with the cloud for aggregation, ensuring a collective and comprehensive global model. The effectiveness of the proposed framework is assessed by comparing its performance metrics with the recent work. The proposed algorithm achieved an accuracy of 99.27 % and stands out by leveraging the concept of collaborative learning. This approach distributes the workload among nodes, relieving the server from excessive burden. Each node is empowered to obtain the best possible model for classification, even if it possesses limited data. This collaborative learning paradigm enhances the overall efficiency and effectiveness of the classification process, ensuring optimal results in scenarios where data availability may be constrained.

Journal Article Type Article
Acceptance Date Nov 24, 2023
Online Publication Date Nov 29, 2023
Publication Date 2024-04
Deposit Date Jan 5, 2024
Publicly Available Date Jan 5, 2024
Journal Internet of Things
Print ISSN 2543-1536
Publisher Elsevier
Peer Reviewed Peer Reviewed
Volume 25
Article Number 101013
Public URL


You might also like

Downloadable Citations