Forest Fire Prediction Using Machine Learning Techniques

Abstract

Forest fires are the most destructive and devastating natural disasters. Forest fire prediction is done to lessen the impact of forest fires in the future. There are several fire detection systems available each with its own strategy. The fire-affected area is forecasted with the help of satellite images. This paper utilizes barometrical factors such as temp, rain, speed, wind, and relative humidity to anticipate the occurrence of a woodland conflagration for fire prediction. The machine learning techniques such as Decision tree, Random forest, Bagging, and Extra tree to solve the fire prediction problem. To prevent overfitting, separate data sets for training and evaluating the model along with cross-validation is performed. Using the Grid Search CV approach, the Decision tree on a range of sub-samples of the dataset is trained and used aggregating to boost projected accuracy to prevent over-fitting. With the proposed model a testing accuracy of 98.36% is achieved for the presented Decision tree based forest fires forecast model. The performance of our hyperparameter tuned model using Grid Search CV performs better compared to existing ML-based model.