Comparison of the Machine Learning Methods to Predict Wildfire Areas
Gözde Bayat, Kazım Yıldız
Abstract
In the last decades, global warming has changed the temperature. It caused an increasing the wildfire in everywhere. Wildfires affect people's social lives, animal lives, and countries' economies. Therefore, new prevention and control mechanisms are required for forest fires. Artificial intelligence and neural networks(NN) have been benefited from in the management of forest fires since the 1990s. Since that time, machine learning (ML) methods have been used in environmental science in various subjects. This study aims to present a performance comparison of ML algorithms applied to predict burned area size. In this paper, different ML algorithms were used to forecast fire size based on various characteristics such as temperature, wind, humidity and precipitation, using records of 512 wildfires that took place in a national park in Northern Portugal. These algorithms are Multilayer perceptron(MLP), Linear regression, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree and Stacking methods. All algorithms have been implemented on the WEKA environment. The results showed that the SVM method has the best predictive ability among all models according to the Mean Absolute Error (MAE) metric.