Clustering of Indonesian forest fires using self organizing maps

Abstract

This paper focuses on clustering the locations of Indonesian forest fires and visualizing them into a two-dimensional map using a self-organizing map (SOM) algorithm. The input data is based on the quantity of the hot spots of forest fires that spread in several locations within ten months period. We analyze the distributions of the hot spots locations of the regions that may have the high frequencies to risk of the forest fires disaster using the SOM algorithm. We have used a principal component analysis (PCA) to reduce the size of the original datasets in order to improve the accuracy of the clustering results. The SOM algorithm has been used to cluster and visualize the map of the hot spots locations into four groups based on the relative similarity of the risks of forest fires on each of the regions such as danger level, low level, high risks, and low risks. From the analysis we have found that a time period where the highest level of quantity and intensity of the forest fires occurs in some regions can be clearly classified.