Assessing the suitability of soft computing approaches for forest fires prediction

Forest fires present one of the main causes of environmental hazards that have many negative results in different aspect of life. Therefore, early prediction, fast detection and rapid action are the key elements for controlling such phenomenon and saving lives. Through this work, 517 different entries were selected at different times for montesinho natural park (MNP) in Portugal to determine the best predictor that has the ability to detect forest fires, The principle component analysis (PCA) was applied to find the critical patterns and particle swarm optimization (PSO) technique was used to segment the fire regions (clusters). In the next stage, five soft computing (SC) Techniques based on neural network were used in parallel to identify the best technique that would potentially give more accurate and optimum results in predicting of forest fires, these techniques namely; cascade correlation network (CCN), multilayer perceptron neural network (MPNN), polynomial neural network (PNN), radial basis function (RBF) and support vector machine (SVM) In the final stage, the predictors and their performance were evaluated based on five quality measures including root mean squared error (RMSE), mean squared error (MSE), relative absolute error (RAE), mean absolute error (MAE) and information gain (IG). The results indicate that SVM technique was more effective and efficient than the RBF, MPNN, PNN and CCN predictors. The results also show that the SVM algorithm provides more precise predictions compared with other predictors with small estimation error. The obtained results confirm that the SVM improves the prediction accuracy and suitable for forest fires prediction compared to other methods.