A comparison of decision tree and backpropagation neural network classifiers for land use classification

Abstract

Decision tree classification techniques have been used for a wide range of classification problems and becoming an increasingly important tool for classification of remotely sensed data. These techniques have substantial advantages for land use classification problems because of there flexibility, nonparametric nature, and ability to handle nonlinear relations between features and classes. This paper compares classification results obtained by using a backpropagation neural network and decision tree classifier. It is shown by a number of studies that neural classifiers depends on a range of user defined factors, that ultimately limits their use. This study show that there are fewer number of user defined factor affecting the accuracy of a decision tree classifier. Further, this study highlight that training a decision tree classifier is much faster and these classifiers are easy to read and interpret as compared to a neural classifier which is a black box. The performance of these two classification system is compared using ETM+ and interferometric SAR data.