Algoritma Backpropagation dalam Akurasi Memprediksi Kemunculan Titik Api (Hotspot) pada Wilayah Kerja Dinas Kehutanan

Abstract

Forest and land fires are an annual disaster issue in Indonesia. The forest area in West Sumatra is ± 2,286,883.10 Ha and 27% or an more than 630,695 Ha of forest area categorized as critical land that has the potential to burn and be damaged. Controlling for forest and land fires in West Sumatra Province was task for Forestry Departement, part of Sumatera Barat Government. One of is task was to reduce the rate of forest destruction. Forest and land fires are an annual disaster issue in Indonesia. The forest area in West Sumatra is ± 2,286,883.10 Ha and 27% or an more than 630,695 Ha of forest area categorized as critical land that has the potential to burn and be damaged. Controlling for forest and land fires in West Sumatra Province was task for Forestry Departement, part of Sumatera Barat Government. One of is task was to reduce the rate of forest destruction. Apart from to extinguishing forest fires directly at the hotspots, preventive action are needed to reduce the possibility of forest and land fires, and one of it is by predicting the possibility hotspots in the future. One of the methods used to predict the possibility hotspots is the use of artificial neural network Backpropagation, this is because Backpropagation has the ability to learn from existing data patterns to calculate the possibility of future events. Data of hotspots that have happened previously and several supporting variables such as air temperature, humidity, rainfall and wind speed, were analyzed and grouped as the basis for the formation of an artificial neural network and for further data training. This learning is done by testing several different network architectures. The results obtained from these tests are the Performance and MSE (Mean Squared Error) values for each network architecture. The test results for each architecture will be compared to determine the best architecture that produces the most accurate predictive value and the smallest MSE. The results of this prediction will later be used as one of the considerations for the Forestry Departement for planning forest and land fire control activities in their area.