Szczegóły publikacji
Opis bibliograficzny
Intelligent fire detection in agriculture using machine learning and embedded systems for risk prevention and improved sustainability / Abdennabi Morchid, Abdennacer Elbasri, Hassan Qjidaa, Rachid El Alami, Badre Bossoufi, Salah Mobayen, Paweł SKRUCH // Scientific Reports [Dokument elektroniczny]. — Czasopismo elektroniczne ; ISSN 2045-2322 . — 2026 — vol. 16 iss. 1 art. no. 9773, s. 1–18. — Wymagania systemowe: Adobe Reader. — Bibliogr. s. 16–17, Abstr. — Publikacja dostępna online od: 2026-02-18
Autorzy (7)
- Morchid Abdennabi
- Elbasri Abdennacer
- Qjidaa Hassan
- El Alami Rachid
- Bossoufi Badre
- Mobayen Saleh
- AGHSkruch Paweł
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 167468 |
|---|---|
| Data dodania do BaDAP | 2026-05-21 |
| Tekst źródłowy | URL |
| DOI | 10.1038/s41598-026-40378-w |
| Rok publikacji | 2026 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Creative Commons | |
| Czasopismo/seria | Scientific Reports |
Abstract
This research proposes the design of an autonomous fire detection system based on a Raspberry Pi 3 B+, combined with smoke and flame sensors, enabling real-time monitoring and increased reactivity to fire hazards. To improve detection accuracy and risk prediction, Machine Learning (ML) algorithms, including Random Forest and Linear Regression, were applied to classify hazard levels and anticipate critical situations. The implementation of these models has enabled accurate classification of risk levels, guaranteeing real-time alerts and rapid decision-making. In addition, anomaly detection techniques based on the Random Forest model have been integrated to identify unusual sensor behavior, ensuring the reliability of the data collected and the correction of any measurement errors. A major contribution of this research lies in the fact that the systems were developed to be adaptable in order to serve areas (i.e., rural and agricultural areas without sufficient Internet access and/or cloud infrastructure) that had little or no access to the Internet. By integrating an embedded, independent, and efficient solution, this system offers a viable alternative to conventional monitoring methods. As part of this research, model performance was evaluated using a confusion matrix for classification accuracy, with the identification of abnormal sensor behavior (anomalies). The performance of each model was evaluated by implementing five-fold stratified cross-validation to confirm their accuracy. The logistic regression model yielded an overall average accuracy of 0.9446 ± 0.0600, with an overall F1 score of 0.9173 ± 0.0744 and a total recall of 0.9250 ± 0.0608. On the other hand, the random forest model produced an overall average accuracy of 0.9860 ± 0.0172, with an overall F1 score of 0.9740 ± 0.0319 and a total recall of 0.9733 ± 0.0327. The random forest demonstrated reliable and balanced classification of fire risk levels compared to the other models in this study. The safety and sustainability of agricultural crops are directly supported by the results of this research, with a reduction in agricultural fire risk through the protection of natural resources and increased resilience of farms through better preparedness for natural disasters. The transition of agriculture to smart systems involves integrated smart approaches to agricultural risk management in order to achieve safer, more sustainable, and more resilient agriculture.
