Riau Forest Fire Prediction using Supervised Machine Learning

Abstract

Forest fire is one of the environmental problems in terms of economically and ecologically detrimental. The number hotspot of forest fires in Indonesia has become increasing dramatically in September 2019 with 16178 hotspots that caused hazardous haze. The hazardous haze has disturbed about 1.04 million and 6.5 million people in Pekanbaru City and Riau Province subsequently. Hence, the development of early warning systems may provide effective strategic information plus accurate prediction results tailored for forest fire prevention and control. Weather data can be used as the main source for analyzing forest fires. This study aims to predict forest fire in Riau, Indonesia. This study has used 1733 weather data in five years (2015-2019). The forest fire prediction models were developed by using two supervised machine-learning techniques namely Decision Tree (DT) and Bayesian Network (BN). The experimental result shows that BN outperforms DT with accuracy rate and RMSE value in pairs of 99.62% and 0.076, and 93.18% and 0.244 subsequently. Despite its low performance, DT able to extract the main factors that caused the forest and land fires efficiently. Thus, it can be concluded that the prediction model using BN has the potential to be used effectively but still has plenty of room for improvement.