ANALYSIS OF OPTIMAL INPUT COMBINATION IN RANDOM FOREST ALGORITHM FOR LAND COVER CLASSIFICATION USING SENTINEL 1 AND SENTINEL 2 SATELLITE IMAGES

• Main Author: Kurnia Kevin Karewur, Dhanny
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76586
• Institution: Institut Teknologi Bandung

Land cover classification has an important role in monitoring the condition of the earth's surface. Conventional field survey methods are not efficient and accurate in collecting land cover information, but remote sensing technology allows monitoring of land cover. One method that has a level of effectiveness in land cover classification with remote sensing data is the random forest method, the random forest method approach can handle the complexity of many features. The combination of random forest model determination can be influenced by the number of features used, the number of decision points, and the ratio of training and test data. The optimal combination of parameters will provide an implementation that has a better level of confidence. In this study, remote sensing imagery used sentinel images 1 and 2. The study found that the model with 11 parameters, 100 decision trees and 70:30 test training ratio achieved a test accuracy rate of 0.764 and a computational time of 0.3 seconds. The applied model was implemented in the research area which produced a kappa coefficient of 0.45 with a sufficient level of confidence (moderate).