Land Cover Classification Assessment Using Decision Trees and Maximum Likelihood Classification Algorithms on Landsat 8 Data

Classification technique on remote sensing images is an effort taken to identify the class of each pixel based on the spectral characteristics of various channels. Traditional classifications such as Maximum Likelihood are based on statistical parameters such as standard deviation and mean, which have a probability model of each pixel in each class. While the object-based classification method, one of which is the Decision Trees, is based on rules for each class with mathematical functions. This study compares the Decision Trees and Maximum Likelihood algorithms for land cover classification in the Surabaya and Bangkalan areas using Landsat 8 data. This research begins with creating Regions of Interest (ROIs) and Rules on images with greater than and less than functions for Decision Trees. The ROIs test was carried out using the Separability Index and matching each class using the Confusion Matrix. The experimental results show that the accuracy value resulting from the Confusion Matrix calculation is 90.48%, with a Kappa Coefficient Value of 0.87. The Decision Trees method produces land cover nigher to the actual condition than the Maximum Likelihood method. The difference in the class distribution of the two ways is not significant. This study is limited because the validation uses manual interpretation results. Future research is expected to use the large-scale classification results from the relevant agencies to verify the classification results and use field data, larger samples of ROIs, and the use of high-resolution imagery in order to improve the classification results.