International Journal of Latest Technology in Engineering, Management & Applied Science (IJLTEMAS)
Volume IX, Issue V, May 2020 | ISSN 2278-2540
Prevention of Forest Fire with AI
Nachiketa Hebbar
Vellore Institute of Technology, Vellore, India
Abstract—Every year forest fires destroy a huge area of forest for fires.However, the problem is, detection can
cover, leaving largescale destruction of flora and fauna in its only be done once a fire is actually started. It is
wake. Forest fires play a major role in driving thousands of also not economically feasible to cover large
species of wildlife to extinction year. Artificial intelligence helps forest covers with cameras and drones.
us predict the future and using it in this domain can successfully
help us predict forest fires and save the wildlife. Any fire Forest Fire Reservoirs: This is simply creating
essentially depends upon 3 factors which are the oxygen, water supplies near forest covers to extinguish
temperature and humidity. This research aims at predicting the fires early. This is system again only works after a
possibility of a forest fire taking place, given the oxygen,
fire takes place and does not help in detection of
humidity and temperature content of a given place. A concept
website that can be created to take inputs from the user and the forest fire
predicts the forest fire probability in real time, is also shown. What our system proposes: Machine learning
I. INTRODUCTION models train on data. So, we take real life
examples of forest fires that took place and collect
F orest or Wildlife fires are uncontrolled fires in area of
combustible vegetation. Depending on the scale of fire it
can be classified as bush fires, forest fires, etc.
the data priorto the fire taking place, which is
publicly available. We have the inputs as oxygen,
humidity, temperature and the output as 0 or 1
They pose a huge risk to wildlife and it becomes pertinent that based on whether or not a fire took place. On
we come up with a solution to counter it .Now the main creating a large enough dataset,we can create a
challenge that comes up here is to detect or predict a wildfire trained machine learning model which can
before it actually happens because once a forest fire gets successfully predict the probability of a fire taking
started it becomes very difficult to put them out before they place in an area given the 3 parameters.
cause large scale irreversible damage. Government can in that sense take necessary
precautions for areas which high probability of a
Machine learning is learning from data to be able to predict fire breaking out.
the future.Hence, we are going to model some parameters
crucial for any forest fire to take place and predict the 2.1 Data Flow Model
possibility of a forest fire taking place based on that. .
1.1. Principle of Wildfire detection Dataset
The detection of a wildfire is primarily dependent upon 3
factors
Oxygen Level: For any fire take place, high oxygen
content is required. So higher the oxygen more is the
probability of a wildfire taking place Machine Learning
Temperature: Obviously for a fire to take place, heat Model trains on the
is favourable. Hence high temperature increases the data.
probability of fire in any region.
Humidity: Obviously Humid weather is unfavourable
for a fire, whereas a dry weather is. Therefore, higher
the humidity, lower the probability of a fire taking Machine learning model
place.
predicts the probability of
II. ALTERNATIVE WORK GOING ON IN THE FIELD AND THEIR fire taking place
DRAWBACK
Camera Surveillance: This approach uses drones
or camera equipment to survey nearby forest cover
www.ijltemas.in Page 45
International Journal of Latest Technology in Engineering, Management & Applied Science (IJLTEMAS)
Volume IX, Is
Issue V, May 2020 | ISSN 2278-2540
2.2 Development of Machine Learning Model Since we obtain the highest accuracy of Logistic
Regression, we opt that model. A brief explanation about
Input: Oxygen, Humidity, Temperature Value Logistic Regression is given:
Output: Probability of Fire Occurrence Logistic Regression:
Dataset: Here is how a sample dataset will look This is a machine learning model that outputs the
like. probability of a particular input instance belonging to
a particular class.
In this case output class are binary: ‘Yes’(A forest
fires likely to take place) , ‘Not’(Forest fire unlikely
to take place).
Hence,wewe can obtain prioritized list of places with
the places with most likely probability of a forest fire
taking place at the top
2.3.1 Large Scale Application
Once we get access to more data the machine
learning model accuracy can be further increased.
increased
On a large scale this can be deployed by all forest
We use a dataset of 100 values for now , but as the scope of authorities so that they have a prioritised list of
the project increases dataset size can also be increased to places with places with maximum likelihood of a fire
achieve higher accuracy. taking place at the top.
This can be combined with web application to give a
2.3 Learning Algorithm nice interface for forest authorities and this provides
This particular problem comes under the category of a way of smarter patrolling so that forests with
Supervised Learning. We train our machine learning mode
model greater likelihood of a fire taking
tak place are given
using the following 3 learning models and compare the maximum patrolling and access to water supply
accuracies: 2.3.2 Web application
1. Linear Regression When this concept is applied and integrated with web
2. Logistic Regression development, we can create a web application that
3. Support Vector Machine simply takes 3 inputs from the user to get the forest
Here is a snippet of the Python code: fire probability.
This can also be used by citizens which will allow
citizens to patrol forests as well and alert higher
authorities in case of a danger.
Here is how a web application created using flask
and html and css looks like:
The following output is obtained using the sklearn library
in python:
Scenario 1: When input is cold weather, low oxygen content
and high humidity(forest fire is not favourable)
www.ijltemas.in Page 46
International Journal of Latest Technology in Engineering, Management & Applied Science (IJLTEMAS)
Volume IX, Is
Issue V, May 2020 | ISSN 2278-2540
III. CONCLUSION
Hence, we can see that machine learning can definitely help
us predict the possibility of a forest fire taking place.After
place
integration of this model with web or app development it can
turn in into application of a large-scale
large use to prevent and
detect wildfires all around the globe.
globe
REFERENCES
[1] V. Kumar, A. Jain, and P. Barwal, “Wireless
“W sensor networks:
security issues, challenges and solutions,” International Journal of
Information and Computation Technology (IJICT),
(IJICT) vol. 4, no. 8,
pp. 859–868, 2014.
[2] M. Hefeeda and M. Bagheri, “Wireless sensor networks for early
detection of forest fires,” in Proceedings of the IEEE International
Internat
Scenario 2:When input is hot weather, high oxygen content Conference on Mobile Adhoc and Sensor Systems,
Systems pp. 1–6, IEEE,
and low humidity. 2007.
[3] S. Eskandari, “A new approach for forest fire risk modeling using
fuzzy AHP and GIS in Hyrcanian forests of Iran,” Arabian
Journal of Geosciences,, vol. 10, no. 8, p. 190, 2017.
[4] D. M. N. Rajkumar, M. Sruthi, and D. V. V. Kumar, “Iot based
smart system for controlling co2 emission,” International Journal
of Scientific Research in Computer Science, Engineering and
Information Technology, vol. 2, no. 2, p. 284, 2017.
[5] Bond W.J., and J.E. Keeley. 2005. Fire as a global ‘herbivore’:
the ecology and evolution of flammable ecosystems. Trends in
Ecology and Evolution 20(7): 387-394.
387
[6] Bratten, F.W. 1969. A mathematical model for computer
allocations
ns of firefighting resources on large wildfires. U.S. For.
Serv. Pac. Southwest For. Range Exp. Stn., Berkeley, CA.
[7] Fried J.S., J.K. Gilless, and J. Spero. 2006. Analysing initial
attack on wildland fires using stochastic simulation.
International Journal of Wildland Fire 15(1): 137-146.
137
www.ijltemas.in Page 47
