Hotspot Classification for Forest Fire Prediction using C5.0 Algorithm

Abstract

Forest and land fires impact the destruction of ecosystems and destroy flora and fauna. Forest fires haze can also disrupt the transportation sector, especially aviation transportation. Forest fire is a recurring disaster problem in Indonesia, especially on Sumatra island. That requires solutions to overcome it, one of which is the monitoring hotspot. A hotspot is an object on the earth's surface represented in a point with certain coordinates that have relatively higher temperatures than its surrounding areas. This study classified hotspots using the C5.0 algorithm to generate forest fire prediction model. The dataset is divided into two categories, namely the explanatory factors representing four region characteristics (cities, river, road, and land cover) and three climate data (rainfall, temperature, and wind speed), and the target class representing the hotspot class (true/false) in the study area, namely Indragiri Hulu Regency, Riau Province, Indonesia. The result is forest fire prediction model that obtained an accuracy of 98.47% on training data, while on test data of 98.68%. The resulting rules are 80 rules excluding three attributes, river, road, and wind speed. The rules can be used as information on preventing forest fires based on the characteristics of the land and the weather of an area.