Classification Performance Analysis of CART and ID3 Decision Tree Classifiers on Remotely Sensed Data

Classification is one of the most commonly employed techniques in remote sensing studies. The thematic maps generated by the RS data classification are employed in a diversity of socio-economic applications. With the advent of space technology, new diverse and advanced RS data are generated at a very high rate. In this study, we implement two decision tree based classification techniques; classification and regression trees (CART) and iterative dichotomizer (ID3), in classifying heterogenous multispectral RS data. The study uses two Landsat-8 study areas; the North Canara District boundary and Kumta Taluk boundary, in Karnataka India. We selected seven level-1 and level-2 LULC classes from Anderson’s (Anderson, A land use and land cover classification system for use with remote sensor data, vol. 964. US Government Printing Office, 1976) classification system for each study area. Class separability is measured between each class pair using the Euclidean distance metric and severely overlapping classes are identified on each data. The paper also discusses different types of decision trees and their attribute selection measures. The results obtained indicate that CART and ID3 are excellent choices for separating severely overlapping spectral class pairs.