A Comparative Study for Predicting Burned Areas of a Forest Fire Using Soft Computing Techniques

Forest fires is an important environmental disasters that have many consequences in our life. As a result, early forecasting and rapid action of fires are needed to control such a phenomenon and saving lives. In this paper, Montesinho Natural Park (MNP) dataset is used to determine the burned areas of a forest fire, three prediction techniques were analyzed; namely Classification and Regression Tree (CART), Multivariate Adaptive Regression Splines (MARS) and Artificial Neural Network (ANN). These algorithms are implemented to specify the technique that would provide best prediction results. The performance of these algorithms was assessed based on three statistical measures, the Mean Absolute Error (MAE), the Mean Squared Error (MSE) and the Root Mean Squared Error (RMSE). From the Statistical measures, it is inferred that MARS algorithm provides the best results in term of performance compared with other methods. The obtained results confirm that MARS improves the forecasting accuracy and has the capability to forecast forest fires effectively with a comparable computational cost.