HOTSPOT PREDICTIVE MODELING USING REGRESSION DECISION TREE ALGORITHM

Abstract

Forest fires had always become an international issue influencing many life sectors, including environmental, social, and economic. The forest fire in 2013 was regarded as one of the worst forest fire tragedies in history, not only in Indonesia but also in the world. Detection of hotspots on the earth's surface by the satellite can be an indication of land and forest fire occurrence. This research aims to build a predictive model of monthly hotspots in Rokan Hilir Regency using the regression tree algorithm. Several variables related to weather information are included, such as rainfall, sea surface temperature, and southern oscillation index. This research used 245 training data and 43 testing data, resulting a predictive model with a correlation of 0.875 and an error rate of 0.166. Based on the values, we can conclude that the performance of the model is considerably good.