Fire Risk Assessment Using Neural Network and Logistic Regression

Forest fire is one of the most important sources of land degradation that lead to deforestation and desertification processes. The presented work describes a methodology that employs logistic regression and artificial neural networks (ANN) to model forest fire risk and to recognize high potential area for fire occurrence. Different satellite and field data have been used in this work to model fire risk. These data include 12 static and dynamic parameters that are effective in fire occurrence and also 2001 to 2004 data was used to create model and data of year 2005 was used to evaluate created model. Two forest fire risk prediction models were created based on logistic regression and neural network in this research and both of them evaluated and compared. The result shows that neural network model is more accurate in fire point classification while logistic regression is sensitive to samples of fire points. To get high accuracy in logistic regression, it is necessary to be equilibrium the proportion of both fire and non-fire samples. Also different neural network structure was tested and the best architecture is a neural network with two hidden layer with 20, 28 neurons and logarithmic-sigmoid transfer function in both hidden layers. Accuracy of logistic regression and ANN in prediction of year 2005 fire was obtained 65.76 and 93.49, respectively.