Perbandingan Algoritma Classification and Regression Tree (CART) dan Random Forest (RF) untuk Klasifikasi Penggunaan Lahan pada Google Earth Engine

Abstract

Land Use and Land Cover (LULC) change are increasing along with the population growth and their activity-followed. Quick and accurate analyses are neded to support the land use management. One of the effective and accurate platforms to generate LULC classification maps are Google Earth Engine (GEE), a cloud-based geospatial by utilising multi-spectral imagery and various processing algorithms.There are at least two LULC classifier methode using the decision tree algorithm: Classification And Regression Tree (CART) and Random Forest (RF). Differences classifier method are allowed to produce difference accuracy level. Few studies have explained how the accuracy level and effectiveness of these two methods.This research aims to identify an accurate and effective LULC classifier method by comparing the accuracy values of the two methods on Landsat 8 and Sentinel 2A imagery. The RF algorithm shows higher accuracy result than the CART algorithm on both Landsat 8 and Sentinel 2A images. The most optimal sample proportion is 60:40 with higher accuracy result on both image and classifier algorithm. The Landsat 8 imagery with sample proportion 60:40 shows the highest accuracy result (OA=93,94 and Kappa=0,927).