Comparative Analysis of Spatial Decision Tree Algorithms for Burned Area of Peatland in Rokan Hilir Riau

Abstract

Over one-year period (March 2013 – March 2014), 58 percent of all detected hotspots in Indonesia are found in Riau Province. According to the data, Rokan Hilir shared the greatest number of hotspots, about 75% hotspots alert occur in peatland areas. This study applied spatial decision tree algorithms to classify classes before burned, burned, and after burned from remote sensed data of peatland area in Kubu and Pasir Limau Kapas subdistrict, Rokan Hilir, Riau. The decision tree algorithm based on spatial autocorrelation is applied by involving Neigborhood Split Autocorrelation Ratio (NSAR) to the information gain of CART algorithm. This spatial decision tree classification method is compared to the conventional decision tree algorithms, namely, Classification and Regression Trees (CART), C5.0, and C4.5 algorithm. The experimental results showed that the C5.0 algorithm generate the most accurate classifier with the accuracy of 99.79%. The implementation of spatial decision tree algorithm succesfuly improve the accuracy of CART algorithm.