AI FOR FOREST FIRE DETECTION
Submitted by
BRITTO RAJ A 920322104009
DIVAKAR S 920322104011
KALISHWARAN 920322104021
MOHAMED FAIZAL U 9203222104029
In partial fulfillment for the award of the degree
BACHELOR OF ENGINEERING IN
COMPUTER SCIENCE AND
ENGINEERING
NM1067 – AI AND GREEN SKILLS
NAAN MUDHALVAN
CHRISTIAN COLLEGE OF ENGINEERING AND
TECHNOLOGY
ODDANCHATRAM - 624619
MAY 2025
Supervised by
K.PRASATH, M.E.
BONAFIDE CERTIFICATE
This is to certify that the project report titled “AI for Forest Fire
Detection ” is the bonafide work of BRITTO RAJ A
(920322104009) , DIVAKAR S (920322104011),
KALISHWARAN S (920322104021), MOHAMED FAIZAL U
(920322104029) who carried out the project work under my
supervision in the Naan Mudhalvan Lab.
SIGNATURE
HEAD OF THE DEPARTMENT SIGNATURE FACULTY
Department of computer science and engineering
CHRISTIAN COLLEGE OF ENGINEERING AND TECHNOLGY
ODDANCHATRAM -624619
INTERNAL EXAMINAR EXTERNAL EXAMINAR
ACKNOWLEDGEMENT
I extend my heart felt gratitude to K.PRASATH, M.E. Co-ordinated of AI and
Green Skills Naan Mudhalvan, for her guidance and support throughout this
project. I also thank my peers and family for their encouragement, without which
this project would not have been possible.
I extend my sincere thanks to all teaching and non-teaching staff of the Department
of Computer Science and Engineering and my peers for their support and
encouragement.
I am deeply grateful to Dr.C.SUNDAR, M.E., Ph.D., Head of the Department, for
her constant support and guidance.
Finally, I thank my family and friends, whose encouragement and patience
motivated me to complete this project successfully.
BRITTO RAJ A 920322104009
DIVAKAR S 920322104011
KALISHWARAN 920322104021
MOHAMED FAIZAL U 9203222104029
INTRODUCTION:
• Forest fires have increasingly become a global concern due to climate
change, deforestation, and human activities. These fires result in the loss of
biodiversity, damage to ecosystems, and significant air pollution.
• Traditional methods of detecting and managing forest fires rely on human
surveillance and satellite imagery, which can delay response time.
• With the evolution of Artificial Intelligence (AI) and data science,
predictive models can be developed to forecast fire occurrences based on
environmental parameters. This project uses AI to enhance forest fire detection by
analyzing key factors like temperature, humidity, wind speed, and rainfall to provide
early warnings and minimize the damage.
ABSTRACTION:
• This project explores the implementation of AI in environmental protection
through a forest fire detection system.
• It uses historical weather and fire data to train a machine learning model
capable of predicting fire risk. The system focuses on building a classifier (for fire/no
fire) or a regression model (predicting burned area) based on input features such as
temperature, humidity, wind speed, and more.
• This model can assist forest departments and disaster management
authorities in deploying preventive measures.
• Additionally, the model can be integrated into real-time monitoring systems
or IoT devices for smart forest surveillance.
DATASET OVERVIEW :
1. Dataset Description:
• Context: Data collected from the northeast region of Portugal,
specifically the Montesinho natural park.
• Purpose: Predict the burned area of forest fires based on
meteorological and other environmental features.
• Size: 517 instances(rows)and 13 attributes(coloumns).
2. Data Preprocessing:
• Handling Categorical Variables: Convert month and day (text) into
numerical form using Label Encoding or One-Hot Encoding.
• Scaling/Normalization: Features like FFMC, DMC, DC, ISI,
temperature, RH, wind, and rain have different scales. Normalize or
standardize them to improve model performance.
PROBLEM STATEMENT :
Brief Overview: The Forest Fires dataset from Montesinho Park, Portugal,
includes 517 instances with 12 input features (like temperature, humidity, wind, rain)
and 1 output (burned area). It mainly aims to predict the size of forest fires based on
meteorological data. Key preprocessing involves encoding categorical data (month,
day), scaling numeric features, handling skewness in area, and detecting outliers. It's
often used for regression or classification tasks.
Key Objectives:
Predict the burned area of forest fires using meteorological and spatial
data. Analyze relationships between weather conditions (temperature, wind, humidity,
rain) and fire severity. Assist in early fire detection and risk management by
identifying patterns. Develop predictive models (like regression or classification) to
anticipate forest fire behavior
METHODOLOGY:
Approach:
• Understand the Problem: Identify that the goal is to predict burned area
based on weather and spatial data.
• Data Preprocessing: Encode month and day variables.
• Exploratory Data Analysis (EDA):Visualize distributions and
correlations.
Detect outliers and trends:
• Model Selection:Try models like Linear Regression, Decision Trees,
Random Forests.
• Model Evaluation:Use metrics like RMSE, MAE, and R².
• Fine-tuning:Optimize hyperparameters to improve accuracy.
Algorithm:
• Linear Regression: For predicting the burned area based on continuous
variables.
• Random Forest Regressor: An ensemble method to improve prediction
accuracy and reduce overfitting.
• Support Vector Regression (SVR): For handling outliers and non-linear
patterns.
Data Collection:
• Use publicly available datasets (UCI/Kaggle).
• Collect real-time data using weather APIs (optional).
Data Preprocessing:
• Remove missing/null values.
• Normalize/scale the data.
• Convert categorical variables using label encoding or one-hot
encoding.
Exploratory Data Analysis (EDA):
• Visualize the distribution of features.
• Analyze correlation between variables.
• Use heatmaps to find strongly related attributes.
Feature Selection:
• Select relevant features using correlation matrix or feature importance
from tree-based models.
Model Building:
• Classification Approach: Logistic Regression, Decision Tree, Random
Forest, SVM, KNN.
• Regression Approach (to predict area): Linear Regression, Random
Forest Regressor, XGBoost.
Deployment (Optional):
• Use Streamlit to create a web interface.
• Deploy model using Flask or Django for backend integration.
FEATURES INCLUDE:
• Temperature (°C)
• Relative Humidity (%)
• Wind Speed (km/h)
• Rain (mm)
• FFMC (Fine Fuel Moisture Code)
• DMC (Duff Moisture Code)
• ISI (Initial Spread Index)
• Fire occurrence (Yes/No or area burned)
Target Variable:
• Fire Occurrence (Binary: Fire / No Fire).
OR
• Area of fire (Regression: in hectares).
RESULTS:
• Model Performance: Metric Example Key Results RMSE (Root
Mean Squared Error) ~20–30 (lower is better). MAE (Mean Absolute Error) ~12–20
(depends on model). R² Score (Coefficient of Determination) ~0.6 to 0.8 (higher is
better).
• Graphs/Visualization: Feature Importance Plot Actual vs Predicted
Plot : Scatter plot comparing true area values vs model predictions. Helps visualize
model accuracy — ideally points align along a 45° line.
DISCUSSION:
INSIGHTS:
• Feature Influence: Meteorological factors like temperature, wind, and
relative humidity are the most significant in determining fire size.
• Model Accuracy: A strong R² score (~0.7–0.8) suggests the model is
capturing important patterns. Residuals Distribution: If residuals are randomly
distributed, the model is likely appropriate for the data.
• Prediction Consistency: The model can predict small and medium-
sized fires well, but might struggle with very large fire areas due to outliers.
Challenges Faced:
• Skewed Target Variable (area): A significant portion of data has zero
or very small values, making it harder to model large fire predictions effectively.
• Feature Correlation: Some meteorological features are highly
correlated, potentially causing multicollinearity and reducing model performance.
• Outliers: Extremely large area values can distort model training and
require special handling (e.g., log transformation).
SOLUTION IMPACT:
• Sustainability impact: The Forest Fire Prediction model can have a
sustainable impact by improving early detection and response to forest fires. Predicting
fire behavior allows for better resource allocation and prevention efforts, minimizing
environmental damage. This contributes to SDG 13 (Climate Action) and SDG 15
(Life on Land) by reducing fire-related risks and protecting biodiversity. By predicting
fire severity, it also helps reduce carbon emissions and wildlife destruction, fostering
long-term sustainability.
• Practical Implementation: The solution can be applied in real-world
settings by integrating the fire prediction model into fire management systems. By
using real-time meteorological data, the model can help predict potential fire outbreaks
and their severity, enabling timely interventions, resource deployment, and preventive
measures. This approach can be used by government agencies, environmental
organizations, and firefighters to reduce damage, protect ecosystems, and mitigate
human impact.
CONCLUSION:
• Summary: The Forest Fires Dataset is used to predict fire sizes based
on meteorological data. Key features include temperature, humidity, and wind, and the
goal is to predict the burned area (area). Data preprocessing involves encoding
categorical variables and scaling numerical features. Models like Random Forest or
XGBoost are used, with evaluation metrics such as RMSE and R². The solution can be
applied for better fire management, contributing to sustainability goals by reducing
environmental impact and aiding in forest conservation efforts.
• Future Work: Integrating Satellite Data: Enhance the model with
real time satellite imagery for more accurate predictions. Expanding Feature Set:
Include more features like soil moisture, vegetation type, or historical fire patterns.
Real-Time Application: Deploy the model into live systems to predict fires and assist
in proactive firefighting strategies.
