Classification Rules for Hotspot Occurrences Using Spatial Entropy-based Decision Tree Algorithm

Abstract

Forest fire is a state where forest affected by fire that led to forest damage and may cause disadvantages in human life. Forest fire event can be monitored using satellite by detecting hotspots as fire indicators at certain times and locations. The purpose of this work is to develop a decision tree to predict hotspot occurrences in Bengkalis district, Riau province Indonesia using the spatial entropy-based decision tree algorithm. The data used are forest fire data in Bengkalis area. The data include city centre, river, road, income source, land cover, population, precipitation, school, temperature, and wind speed. The results of this work using the 5-fold cross validation test are decision trees with the average accuracy of 89.04% on the training set and 52.05% on the testing set. The tree has 560 nodes with the land cover layer as the root node. From the decision tree, as many 255 rules were obtained to classify hotspot occurrences.