Flame Prediction: Using Machine Learning to Tackle Forest Fires

Forests play an important role in environmental preservation. Fires occurring in forest are becoming more common day by day which is a major threat of lives of planet and people. Quick action on these forest fires can saves a lot of lives and vegetation and can eventually help minimise any further harm. In this research paper, various machine learning models are applied, including
$K$
Nearest Neighbour, Naive Bayes and Decision tree, in the Maharashtra region of India. In the models training and testing various parameters and features are taken into consideration including temperature, humidity, wind, precipitation, rainfall and the account of fire being present or not in that specific region. The data and the results show that the Decision tree model gave the highest accuracy of
$96.32{{\% }}$
, followed by the KNN classifier at
$95.38{{\% }}$
and the Naive Bayes classifier at
$93.91{{\% }}$
. The Area Under the Curve (AUC) scores for the models were as follows: The KNN model scored 0.85, the Naive Bayes model scored 0.73, and the Decision Tree model also scored 0.85. According to the findings, Decision Tree is the most effective algorithm for predicting forest fires in Maharashtra. The study develops a framework for predicting forest fires using machine learning algorithms that may be used in different regions and countries. The findings can assist forest management authorities in taking preventive measures and reducing the severity of forest fires.