Comparison of supervised algorithms for built-up classification in Indonesia metropolitan

Built-up area are covering less than one percent but are estimated grow rapidly due to human population and economy activities. The built-up area information is crucial for measuring achievement of Sustainable Development Goals (SDGs) number 11 in.making sustainable cities. Supervised machine learning from public satellite imagery, such as Landsat 8 and VIIRS-DNB, is one of method to achieve that information. In this paper, comparison of supervised algorithms is conducted to measure the performance and the ability to classify the built-up class and non built-up. Algorithms used here are Support Vector Machine (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost). The research focused on distinguish built-up area in six metropolitan area of Indonesia. Parameter tuning are conducted to get the best model of every algorithm in each area. Using Area Under Receiver Operating Characteristic (AUROC) as comparison indicator, XGBoost outperformed other algorithms in five metropolitan areas. Using best model in each area, built-up area classification was conducted in 2014 and mostly built-up area was shown accumulated in center. Then, built-up transformation was measured from 2015 to 2019 and found that Jabodetabek and Kedungsepur experienced the larger built-up area transformation compared the others.