Machine Learning Algorithms Based Models for Forest Fires Prediction: An Analytic and Comparison Study

This paper describes the outcomes of an investigation of several machine learning algorithms as models for forest fires prediction. The conducted comparison and analytic study is based on the performance evaluation of the employed algorithms in terms of accuracy, recall and precision. The aim of this study being the identification of the most efficient ML algorithm based model for forest fires prediction. We have considered in this study four datasets. The prediction is based on the meteorological data corresponding to the critical weather elements that influence the forest fires occurrence (namely temperature, relative humidity, wind speed and rain) and the fire weather index system components. This study revealed that the naive Bayesian and the decision tree algorithms perform well for all datasets, thus they are suitable for our purpose since they give tradeoff between performance (in terms of accuracy, recall and precision) and the model build time. This research provides an orientation for the selection of the most suitable ML algorithm as model for forest fires prediction in Algeria.