A Hybrid of Seasonal Autoregressive Integrated Moving Average (SARIMA) and Decision Tree for Drought Forecasting

Abstract

Drought is one of the triggers for forest fires due to depletion of surface water reserves. Along with the frequent drought, the incidence of forest fires has also increased. Therefore, it is important to know or forecast drought to take precautions. In this study, drought forecasting was carried out by applying the concept of data mining classification methods and forecasting methods. This classification uses the decision tree (CART) method, which is a method that aims to see the rules resulting from the classification of existing data. While forecasting uses the SARIMA method, this method is used to predict the factors that cause drought (temperature, humidity, and rainfall). Furthermore, the rule of the classification results is used to classify the results of forecasts. Based on the implementation of the CART algorithm which is evaluated using a confusion matrix is able to achieve an accuracy of 91.33%. Based on the implementation of the SARIMA method, a model is obtained for each variable to build forecasting. Each model was selected based on AIC criteria, and evaluated using MSE. The optimal model for temperature (Tx) is SARIMA (1, 1, 0) x (0, 1, 1, 12) with the MSE value of 0.15. While the selected model for humidity (RH_avg) is SARIMA (0, 1, 1) x (1, 1, 1, 12) with the MSE value of 3.85, and the optimal model for rainfall (RR) is SARIMA (0, 1, 1) x (0, 1, 1, 12) with the MSE value of 8.61.