Application of machine learning and neural networks in fire data analysis

Forest fires have been regarded as one of the most important environmental concerns for long because they can cause a great damage to forest ecosystems and human lives, and the predictions for forest fires have been implemented in many ways. This paper responds to the challenges of predicting forest fires by using various and inclusive factors implemented in multiple machine learning algorithms and neural networks to do the classification and linear regression. The record of forest fires in Montesinho natural park in Portugal will be chosen as a sample. In order to maximize the performance of the classification algorithm and model accuracy, dividing the sample data to 0 hectare burned area, greater than 0 and less than 17 hectares burned area, and greater than 17 hectares burned area is the most reasonable. To improve the accuracy of the regression methods, pre-processing and cross-validation are used. The purpose of the research is to help firefighters predict fire conditions based on climate data and to test the performance of machine learning algorithms and neural networks on a forest fire dataset.