Estimation the expected loss from a forest fire related to geographical and meteorological parameters by using artificial intelligence methods

Forest fires are environmental issues that create economic problems as well as ecological damage. In this study, the loss due to the possible forest fire was estimated by using the records of previously occurred forest fire parameters like geographical conditions of the existing environment, day and hour when the fire broke out, meteorological data like temperature, humidity and wind speed, and type of trees and number of trees in unit area. The estimated output of this study is the lost area due to the fire. With reference to this output, an analysis about how a forest fire affects the environment was studied and a guideline for providing the proper resource management of fire fighting was aimed. The data used in this analysis was obtained from the Department of Forestry, a section of the Ministry of Forestry and Water Affairs in Turkey. This data set consists of 10.597 forest fire records which belong to the years between 2000 and 2009. During the organization of data, different fuzzy models were created for input parameters. The tests were implemented by using Multilayer Perceptron, Radial Basis Function Networks (RBFN), Support Vector Machines (SVM) and fuzzy logic. In order to observe improvements in the test results, input parameters were reorganized by using Principal Component Analysis (PCA). To create a reliable prediction model different outputs were analyzed. As a result, for performance measurement values were obtained as MSE 53,68 hectare², MAE 3,36 hectare, MAPE 51%, RMSE 7,33 hectare. From fuzzy logic experiments, the best total estimate rate was calculated as 99,94% and the best average accuracy value for fuzzy groups was obtained as 65,63%.