Power Predictive Modeling for Fire Hazards Using Machine Learning

Abstract

This paper presents a fire prediction protocol that integrates deep learning techniques with Internet of Things (IoT) infrastructure for rapid fire detection and enhanced safety measures. The proposed system utilizes an optimized YOLO model for real-time flame detection by processing video and images based on color, temperature, and shape analysis, achieving an accuracy of up to 81% in fire hazard detection. It incorporates flame and smoke sensors for hazard detection, triggering local alerts via a buzzer and LED, and providing remote warnings to users through a Telegram bot. The system architecture encompasses sensor selection, design, and hardware/software implementation using the Arduino IDE. Test results demonstrate improved response time and notification reliability, validating the system\'s efficiency in enhancing fire safety.

Introduction

Purpose and Motivation:

Traditional fire detection systems face delays and lack remote monitoring. This project aims to create a real-time, IoT-based fire detection system using flame and MQ2 smoke sensors, enhanced by deep learning (CNNs and YOLO models) and Telegram alerts, improving both detection speed and accessibility.

? Key Objectives:

1. Develop a hybrid fire detection system combining shallow (color models) and deep learning (CNNs, YOLO).

2. Integrate IoT components (ESP8266, MQ2, flame sensor) for real-time detection and alerts.

3. Enable remote and local notifications (Telegram, buzzer, LED).

4. Achieve faster and more accurate detection for both residential and industrial use.

???? Techniques and Technologies:

1. Machine Learning Approaches:

• Shallow learning: RGB, HSI, YCbCr color models for flame detection.

• Deep learning:

  • CNNs for image classification.

  • YOLOv5 model trained on 172 flame images (augmented to 1,720).

  • Achieved 81% accuracy in flame detection.

2. IoT Integration:

• Flame and smoke sensors connected to ESP8266 microcontroller.

• Real-time alerts via Telegram bot.

• Local alerts through LED and buzzer.

???? Literature Survey Highlights:

• Color models and CNNs are effective, with accuracy over 97–99%.

• IoT integration (MQ2 sensors, real-time data) is widely adopted.

• Edge computing and sensor fusion offer further improvements in latency and accuracy.

• Challenges include computational costs, environmental noise, and dataset limitations.

????? System Requirements:

Functional:

• Real-time fire/smoke detection.

• Local and remote alerting.

• Integration of multiple sensors (flame, gas).

Non-functional:

• 95% detection accuracy.

• Low latency.

• Robust Telegram bot for notifications.

Hardware:

• Flame sensor, MQ2 gas sensor, ESP8266, buzzer, LED.

Software:

• Arduino IDE, Python 3.9+, TensorFlow/Keras, OpenCV, Telegram API.

???? System Design and Implementation:

1. Shallow Learning: Uses RGB and HSI color spaces.

2. Deep Learning: Optimized YOLOv5 for flame detection.

3. IoT System:

   • Real-time monitoring with ESP8266.

   • Local alarm system.

   • Instant Telegram notifications.

???? Performance Results:

• Model accuracy: 81%.

• Real-time alerts and monitoring successfully implemented.

• Balanced trade-off between shallow and deep learning for efficient fire detection.

Conclusion

The Machine Learning Driven Protocol for Fire Prediction project successfully integrated hardwarebased IoT implementations and advanced machine learning techniques to develop efficient and scalable fire detection and alerting system. The system effectively combines real time data collection from sensors with machine learning algorithms for fire prediction and a Telegram bot for immediate user notifications. The project explored both shallow and deep learning methods, with deep learning showing superior performance. The integration of IoT and machine learning provides a synergistic approach to address realworld fire detection challenges.

References

[1] Demirel, Hasan. (2009). Fire detection in video sequences using a generic color model. Fire Safety Journal. 44. 147-158. 10.1016/j.firesaf.2008.05.005. [2] Shen, Dongqing& Chen, Xin& Nguyen, Minh & Yan, Wei. (2018). Flame detection using deep learning. 416-420. 10.1109/IC- CAR.2018.8384711. [3] Agarwal, T. (2019, August 12). MQ2 Gas Sensor - Working Principle & its Applications. ElProCus - Electronic Projects for Engineering Students. https://www.elprocus.com/anintroductionto mq2gassensor [4] Qureshi, Waqar&Ekpanyapong, Mongkol& Dailey, Matthew &Rinsurongkawong, Suchet&Malenichev, Anton &Krasotkina, Olga. (2015). QuickBlaze: Early Fire Detection Using a Combined Video Processing Approach. Fire Technology. 52. 10.1007/s10694-015-0489- 7. [5] Khan F, Xu Z, Sun J, Khan FM, Ahmed A, Zhao Y. Recent Advances in Sensors for Fire Detection. Sensors (Basel). 2022 Apr 26;22(9):3310. doi: 10.3390/s22093310. PMID: 35590999; PMCID: PMC9100504. [6] Lv J, Qi Y, Huang L, Zong K, Du W. A fire detection system based on one-chip computer. Journal of Computational Methods in Sciences and Engineering. 2024;24(4-5):2227-2239. doi:10.3233/JCM-247254 [7] Cetin, A. E., Dimitropoulos, K., Gouverneur, B., Grammalidis, N., Gu¨nay, O., Habibo?glu, Y. H., To¨reyin, B. U., &Verstockt, S.(n.d.). Video fire detection – Review. Digital Signal Processing, 23(6), 1827–1843. https://doi.org/10.1016/j.dsp.2013.07.003 [8] Khan, F., Xu, Z., Sun, J., Khan, F. M., Ahmed, A., & Zhao, Y. (2022). Recent advances in sensors for fire detection. Sensors, 22(9), 3310. https://doi.org/10.3390/s22093310 [9] Shen, D., Chen, X., Nguyen, M., & Yan, W. Q. (2018). Flame detection using deep learning. Flame Detection Using Deep Learning, 416–420. https://doi.org/10.1109/iccar.2018.8384711 [10] C¸ elik, T., &Demirel, H. (n.d.). Fire detection in video sequences using a generic color model. Fire Safety Journal, 44(2), 147–158. https://doi.org/10.1016/j.firesaf.2008.05.005

Copyright

Copyright © 2025 Rashika M E. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.